Wishful Thinking or E�ective Threat?
Tightening Bank Resolution Regimes and Bank Risk-Taking
Magdalena Ignatowski and Josef Korte∗
Draft, April, 2013
Abstract
We propose a framework for testing the e�ects of changes in bank resolution regimes on bank
behavior, particularly on a variety of risk- and business model-measures. By exploiting the di�er-
ential relevance of recent changes in U.S. bank resolution law (i.e. the introduction of the Orderly
Liquidation Authority, OLA) for di�erent types of banks, we are able to simulate a quasi-natural
experiment to test otherwise endogenous e�ects in a di�erence-in-di�erence framework. To the
best of our knowledge, this identi�cation strategy is unique in its application to regulatory changes
in bank resolution. To test our hypotheses, we use a three level dataset: Holding aggregates, bank
level data, and loan level data. We �nd banks that are more a�ected by the introduction of the
OLA to signi�cantly decrease their overall risk-taking and to shift their business model and new
loan origination towards lower risk - indicating the overall e�ectiveness of the regime change. This
e�ect, however, does not hold for the largest and most systemically important institutions, indi-
cating that the application of the OLA does not represent a credible threat to these institutions,
leaving the too-big-to-fail problem unresolved. Finally, we �nd no evidence of gambling between
the announcement and enactment of the OLA, presumably since the legislation was passed compa-
rably quickly. Our results intend to contribute to the emerging literature evaluating implications
of new regulatory policies, and allow relevant conclusions for the design of bank resolution law,
e.g. in the context of the European Banking Union.
JEL classi�cation: G21, G28, G33
Keywords: Bank resolution, bank insolvency, Orderly Liquidation Authority, FDIC, bank behavior,
risk-taking
∗Goethe University Frankfurt, Faculty of Economics and Business Administration, Grueneburgplatz 1, 60323 Frank-furt am Main, Germany, Email: ignatowski@�nance.uni-frankfurt.de and josef.korte@�nance.uni-frankfurt.de
1
Prelude
On June 30, 2010, bank resolution law - under which the Federal Deposit Insurance Corporation (FDIC)
was able to close any insured depository institution in the U.S. - was applicable to about 10.9% of the
Goldman Sachs Group's subsidiaries. At the end of the next reporting quarter, the FDIC had been
enabled by the U.S. Congress to eventually resolve 100% of Goldman Sachs Group or any Finan-
cial Holding Company according to an extension in bank insolvency law termed Orderly Liquidation
Authority (OLA) that had been introduced as part of the Dodd-Frank Act (DFA).
At the time, the Financial Times applauded that �the Dodd-Frank bill makes important strides in
ending government guarantees [...] and disincentivising risky behaviour. [...] In place of government
bail-outs (like AIG) and painful bankruptcies (like Lehman Brothers) comes a new 'Orderly Liquidation
Authority�'.1 And the Economist concluded that �the most important provision [of Dodd-Frank] is
the resolution authority under which federal regulators can seize any �nancial company [...]. This
is an improvement on the status quo.�2 Have these expectations come true? Did such a dramatic
change in bank resolution powers in�uence bank behavior, risk-taking, and business model choices? Do
banks operate di�erently when the threat of being resolved and liquidated by the regulator becomes more
realistic in legal, operational, and �nancial terms?
1 Introduction
When governments were confronted with seriously distressed banks during the global �nancial crisis
of 2008 and the subsequent European sovereign debt crisis, existing resolution tools proved mostly
inappropriate, either because they did not take into account distinctive features of banks or authorities
lacked to some extent empowerment, �nancial resources, and cross-border cooperation to e�ectively
resolve failed banks. A comparison of the failure resolution of Lehman Brothers and Washington
Mutual in September 2008 illustrates the importance of e�ective and appropriate bank resolution
mechanisms.3 Following these recent crisis experiences, bank regulators and legislators have discussed
and brought into force signi�cant changes to bank resolution regimes4 in an e�ort to improve future
bank failure resolution and ultimately to prevent future crises (e.g. Dodd-Frank Act in 2010, German
Bank Restructuring Act in 2011, and Financial Stability Board in 2011).
E�ective and enforceable bank resolution mechanisms are not only of vital importance to deal
with failing banks and minimize costs associated with bank failures. Beyond that, they can have
a disciplining e�ect and thus reduce the probability of bank failure ex ante. Bagehot (1873) already
pointed to moral hazard and excessive risk-taking induced by banks' expectation for bailout. Although
various rationales for bailout policies can be formulated (e.g. Acharya and Yorulmazer (2007); Diamond
and Dybvig (1983); Diamond and Rajan (2005)), several recent studies provide empirical evidence for
1See Financial Times, July 12, 2010.2See The Economist, July 3, 2010.3When Lehman Brothers �led for Chapter 11 bankruptcy protection on September 15, 2008, the bankruptcy �ling
constituted a default action in derivative contracts, leading to massive terminations of derivative positions. As LehmanBrothers was not allowed to provide liquidity to its subsidiaries, its foreign legal entities entered bankruptcy proceedingsas well. At the time of Lehman Brother's failure, Washington Mutual experienced a bank run and was put into FederalDeposit Insurance Corporation (FDIC) receivership by its regulator, the O�ce of Thrift Supervision (OTS), on September25, 2008. FDIC sold Washington Mutual's assets, deposit liabilities and secured debt immediately to JPMorgan Chaseand the remaining holding company �led for bankruptcy protection the next day. Although Washington Mutual'sbusiness had been materially di�erent from Lehman Brothers', its banking business continued to operate without majorinterruptions, unlike the failure of Lehman Brothers. FDIC (2011) provides an extensive discussion of the di�erencesbetween Lehman Brother's bankruptcy under Chapter 11 and a hypothetical resolution under a special bank resolutionregime, i.e. the Orderly Liquidation Authority.
4We use the term 'bank resolution regime' in a wide meaning, not just refering to the actual legal provisions, but alsoto the (�nancial or operational) empowerment of resolution authorities. Also, with regard to a�ected institutions, wedo not just refer to banks in their form as insured deposit-taking intermediaries, but to �nancial institutions with bankfeatures in general (e.g. �nancial or bank holding companies).
2
the moral hazard e�ect of bailout (expectations) on risk-taking, e.g. Black and Hazelwood (2012);
Dam and Koetter (2012); Duchin and Sosyura (2012). Reversely, when bailout guarantees cease to be
implicit through a credible and enforceable improvement in bank resolution regimes, we expect banks
to change their behavior towards less risk-taking and lower probability of distress. This hypothesis is
proposed in a recent model of DeYoung et al. (2013), which suggests that a credible improvement in
resolution regimes can increase overall bank discipline. We take this as a theoretical foundation for
our empirical evaluation.
The introduction of the Orderly Liquidation Authority provides an ideal setup to study this disci-
plining e�ect on bank behavior. The OLA has been established through the Dodd-Frank Wall Street
Reform and Consumer Protection Act of 2010 (DFA) and enables the FDIC to seize control and liq-
uidate any �nancial institution in distress through its administrative resolution regime. Before DFA's
enactment, the FDIC's resolution authority only comprised insured depository institutions. With
the OLA, FDIC's authority has been extended to institutions that were previously exempted from
any speci�c bank resolution regime, namely Bank Holding Companies (BHCs) and non-bank �nancial
companies. In this paper, we distinguish between BHCs with large nonbank �nancial asset holdings
on the one hand and BHCs with mainly depository bank holdings and independent banks on the other
hand. By exploiting the di�erential relevance of the OLA to these groups, we are able to simulate a
quasi-natural experiment that allows us to test otherwise endogenous e�ects in a di�erence-in-di�erence
framework.
We address a series of important and novel questions in this paper: Do banks change their behavior
when bailout expectations vanish and the threat of being resolved in case of failure becomes more
realistic? More precisely: Is the OLA a credible and e�ective improvement to the resolution regime
leading to a reduction in return volatility, asset risk, and default probability of a�ected institutions? Do
banks adjust their business models following the OLA, e.g. with regard to their securities investments,
trading acitivities, or funding structure? Is there a change in risk-taking when it comes to new business,
more speci�cally do banks approve and originate less risky mortgage loans? Is the improvement in the
resolution regime e�ective for all banks, and is the resolution threat credible and e�ective even for banks
that are deemed 'too-big-to-fail'? Finally: Can we observe a reverse e�ect between announcement and
enactment of the resolution policy change, which would correspond to theories on gambling (compare
e.g. Fischer et al. (2012); Murdock et al. (2000))?
These questions are addressed using a three level dataset: Holding aggregates, bank level data, and
loan level data. We �nd banks that are more a�ected by the introduction of the Orderly Liquidation
Authority to signi�cantly decrease their overall risk-taking after the OLA becomes e�ective relative
to the control group of non-a�ected banks. On a more detailed level, we �nd that a�ected banks also
shift their business model and new loan origination towards lower risk. Our results indicate the overall
e�ectiveness of the regime change, which can indeed be interpreted as an improvement in available
resolution technology. However, the overall e�ect does not hold for the largest and most systemically
important institutions, indicating that the application of the OLA does not represent a credible threat
to these institutions. Hence, even the introduction of the OLA seems to leave the 'too-big-to-fail'
problem unresolved - at least for the very largest banks. Finally, we �nd no evidence of gambling
around the announcement and enactment of the OLA, presumably since the legislation was passed
and enacted comparably quickly.
We focus our analysis on the U.S. due to the unique identi�cation opportunity and due to data
availability, but our results have wider implications. They are not only of concern in evaluating the
e�ectiveness of resolution policy change in the U.S., but can also contribute to regulatory discussions
in the context of, for example, an EU-wide joint bank recovery and resolution policy framework that
is proposed as part of the planned European Banking Union (European Commission, 2012).
3
Our paper is intended to contribute to the latest literature on the e�ects of regulatory actions
on bank behavior, particularly risk-taking, e.g. Berger et al. (2012); Black and Hazelwood (2012);
Dam and Koetter (2012); Duchin and Sosyura (2012). While these papers mainly focus on the ef-
fects of government bailout policies, we investigate the e�ects of an ex ante disciplining regulatory
approach. Although an economic rationale for such disciplining resolution policies has been modeled
before (Acharya, 2009; Acharya and Yorulmazer, 2008; Perotti and Suarez, 2002), empirical evidence
is limited to the (non-)application of resolution rules by regulators (Brown and Dinç, 2011; Kasa and
Spiegel, 2008). A vital implication of resolution regimes, however, has so far mostly gone unevalu-
ated: the e�ects of their tightening on bank behavior. Therefore, this paper is intended to provide an
empirical test of the credibility and e�ectiveness of changes in resolution regimes with regard to their
implications for bank behavior. As a methodological contribution, we propose an identi�cation setup
that is - to the best of our knowledge - novel to testing the e�ects of changes in resolution regimes.
The remainder of this paper is organized as follows. Section 2 gives an overview of the related
theoretical literature and core �ndings of existing empirical research. Our key hypotheses are proposed
against this background. In Section 3, we introduce our identi�cation strategy and present initial
indicative evidence. Our full model and dataset is described in Section 4. Section 5 presents the results
of the analysis, several extensions, and is complemented with robustness tests. Section 6 concludes
and provides policy implications.
2 Background, related literature and key hypotheses
2.1 How regulation drives bank risk-taking
Financial economics literature has identi�ed several determinants for bank risk-taking, among them the
degree of competition, the degree of information transparency on bank risks, and ownership structure,
but also incentives created by bank regulation and safety nets. In this section, we revisit theoretical and
empirical literature to investigate how regulation � and particularly the resolution regime � interacts
with bank risk-taking.
In general, the literature has mostly focused on four main forms of bank regulation: deposit in-
surance, capital regulation, restrictions on bank activities, and resolution of banks. Deposit insurance
schemes are often described as safety nets against bank runs. However, deposit insurance at a �xed
rate (independent of the risk of banks' assets) creates a moral hazard problem, as banks can borrow
funds cheaply through insured deposits and invest them in risky assets (Kareken and Wallace, 1978;
Merton, 1977). Moreover, insured depositors have little incentive to monitor the bank.5 This moral
hazard problem can be mitigated by making the deposit insurance explicit and leaving some credi-
tors uninsured (Calomiris, 1999; Gropp and Vesala, 2004). Other design features of deposit insurance
such as funding, premium structure or membership requirements can also alleviate the moral hazard
problem (Barth et al., 2004).
The purpose of capital regulations is to reduce banks' � more precisely bank owners' � risk-taking
incentives through forcing them to leave some of their capital at risk as a bu�er for future losses.
However, a simple capital-to-asset ratio provides incentives to shift to riskier asset portfolios, thus
increasing risk-taking behavior (Koehn and Santomero, 1980). A risk-based capital ratio that accounts
for asset quality can reduce this asset-substitution problem (Kim and Santomero, 1988; Repullo, 2004).
Yet, several further theories suggest negative e�ects of capital regulation on bank behavior.6
5Empirical cross-country studies strongly con�rm the moral hazard incentive of deposit insurance.Demirgüç-Kunt andDetragiache (2002) show that the existence of deposit insurance increases the probability of banking crises and that thise�ect is even stronger the more coverage the deposit insurance provides. Demirgüç-Kunt and Huizinga (2004) provideevidence for the adverse e�ect of explicit deposit insurance on market discipline.
6Capital regulations might actually increase bank risk-taking (Besanko and Kanatas, 1996; Blum, 1999; Murdock et al.,
4
Similar to capital regulation, restrictions on bank activities also aim at more prudent risk behavior
by restraining banks from engaging in other risky businesses outside their original activities (Boyd et al.,
1998). Empirical studies provide mixed evidence on the risk mitigating e�ect of activity restrictions
(e.g. Barth et al. (2004)).
Resolution of distressed banks is probably the most intricate regulatory area regarding risk-taking
incentives. Overall, there are two (opposing) regulatory approaches to handling a distressed bank:
bailing out the bank in order to preserve it as a going concern and resolving the bank either through
acquisition by another �nancial institution (i.e. purchase and assumption) or straightforward closure
and liquidation. One line of theory predicts that the expectation of being bailed out increases banks'
moral hazard as creditors anticipate loss protection in case of bank failure and have little incentives to
monitor the bank (or to adjust risk premiums as indicated in Sironi (2003) and Gropp et al. (2006)). A
di�erent theoretical approach suggests that bailout guarantees can increase charter values (i.e. through
lower funding cost) and hence decrease incentives for excessive risk-taking as banks fear losing these
charter values (Keeley, 1990). Connecting both theories, Cordella and Yeyati (2003) and Hakenes
and Schnabel (2010) develop models where the positive charter value e�ect can actually outweigh the
negative moral hazard e�ect and thus lead to more prudent risk-taking behavior of banks protected
through bailout guarantees. However, their models depend on speci�c economic circumstances, banking
sector characteristics and/or bailout policy designs. Empirical evidence tends to support the view that
bailout policies rather increase than decrease bank risk-taking and moral hazard in the long run.7
A credible resolution threat of closing or selling banks in case of failure should decrease excessive
risk-taking incentives ex ante. Theoretical models however predict certain caveats: According to Davies
and McManus (1991), the e�ect of the closure threat on bank risk-taking depends on the bank's
'healthiness' (i.e. capital base) and the regulator's closure rule (i.e. specifying closure at a certain
capital level). Mailath and Mester (1994) model a time-inconsistency problem where the regulator's
bank closure decisions interact with banks' asset choices, leaving the regulator unable to credibly
commit to closure policies. Apart from ex ante incentives, closing or selling banks in case of failure can
also impact ex post incentives of surviving banks. Perotti and Suarez (2002) consider a model where
the acquisition of failed banks enhances charter values of surviving banks (i.e. through greater market
concentration) and thus increases surviving banks' incentives for prudent risk behavior. Another
conceivable implication on bank behavior could be 'gambling for resurrection'. As theoretically shown
in Murdock et al. (2000), bank's incentive to gamble increase, when they lose their charter values. The
withdrawal of an (implicit) bailout guarantee due to an introduction of a credible resolution threat
can imply higher funding cost and thus a loss in charter values. Hence, banks might start gambling as
a reaction to a change in resolution policy.
Taken together, the existing literature proposes, models, and evaluates several e�ects of bank failure
resolution (bailout or closure) on bank behavior. To the best of our knowledge, however, there has not
been any study so far that empirically investigates the e�ects of tightening resolution regimes on bank
risk-taking.
2000) and decrease lending activity (Thakor, 1996). Moreover, the recent �nancial crisis revealed the shortcomings ofrisk-based capital regulation: Neither did they score well on predicting failure (Berger and Bouwman, 2012; Blundell-Wignall and Atkinson, 2010), nor did they prevent regulatory arbitrage. Rather, when the risk of certain assets isnot properly estimated by a regulator, banks have strong incentives to acquire and hoard these assets, thus increasingsystemic risk (Acharya et al., 2013).
7Black and Hazelwood (2012)and Duchin and Sosyura (2012) provide evidence that (at least large) TARP-funded U.S.banks increased risk-taking after the capital injection. Dam and Koetter (2012) exploit a dataset on capital injectionsin Germany and �nd that bailout expectations (through observed capital injections) increase risk-taking in the wholebanking sector (measured as probability of default). However, using the same dataset, Berger et al. (2012) show thatbanks receiving capital injections decrease risk-taking (measured as ratio of risk-weighted assets to total assets). Theresults in Gropp et al. (2011) are also mixed: They �nd no evidence for increased risk-taking by banks protected bybailout guarantees.
5
2.2 A theoretical model of bank closure
In the theoretical literature, bank resolution regimes have attracted more and more interest over
recent years. One of the most comprehensive theoretical models of the interaction between resolution
law, its credibility and application, and bank behavior was recently o�ered by DeYoung et al. (2013).
Building on the time-inconsistency problem of bank closure decisions formulated by Mailath and Mester
(1994) and Acharya and Yorulmazer (2007), the authors model the regulatory closure of a bank as a
trade-o� between short-term liquidity and long-term discipline. The model assumes banks that are
inherently fragile and su�er from moral hazard with regard to excessive risk, complexity, and volatility.
Essentially, there are two alternatives for the regulator to deal with this. On the one hand, banks can
be disciplined by a strict closure and resolution policy in case of failure. Unfortunately, this discipline
only materializes in the long run. On the other hand, while they help to establish discipline, available
resolution technologies usually su�er from limitations. These limitations, such as slow processes,
missing information, or legal limits to available regulatory instruments, might (temporarily) lead to
illiquidity in the case of bank closures. This might result in a detrimental impact on the economy as
a whole (e.g. Ashcraft (2005)). Hence, the regulator - despite knowing about the long run bene�ts
of discipline - also has an intrinsic motivation to prefer bailouts or forbearance over straightforward
closure.
DeYoung et al. (2013) model the outcome of this trade-o� as determined by two parameters. The
�rst one is the time discount rate of the regulator - the higher it is, the stronger is the regulator's
preference for liquidity, i.e. bailout. E�ectively, this discount rate proxies for the pressure for imme-
diacy that regulators and economic policy makers are experiencing, e.g. political pressure to preserve
liquidity during a crisis.8 The resolution technology available to the regulator is the second parameter
determining the trade-o�. The better this technology is, the faster and more e�cient a bank closure
can be executed, the more liquidity is preserved. Consequently, regulators with better resolution tech-
nologies at hand are - under the assumption of equal time discount rate - more induced to enforce
discipline, i.e. closure.
This model provides several testable implications. First, improvements in resolution technology,
such as legal changes or operational empowerment of the regulator, make a regulatory policy preferring
discipline (i.e. closure in case of failure) more likely. If the technological improvement is known and
credible to banks, they will act rationally by adjusting their behavior towards more discipline ex
ante. Hence, an improvement in resolution technology should induce less excessive risk-taking and
the adoption of more conservative business models, ceteris paribus. Second, this outcome depends on
the credibility of the application of the new resolution technology. The new policy instruments will
only be e�ective when complemented by political will, i.e. a low time discount rate that increases
the willingness of regulators to accept potential short-term illiquidity following bank resolution for
long-term gains in discipline. Using these general implications as our theoretical foundation, we test
whether the change in speci�c resolution technologies is indeed an e�ective and credible improvement
that alters the behavior of a�ected banks.
2.3 Hypotheses on the e�ects of tightening resolution regimes
Building on the theory of bank resolution and previous �ndings discussed above, we yield the following
hypotheses and subject them to econometric testing.
8Several empirical studies con�rm the tendency for bailout and forbearance in times of macroeconomic or systemicstress. Brown and Dinç (2011) and Kasa and Spiegel (2008), for example, �nd that regulators are less likely to close abank if the whole banking system is in a crisis.
6
Main hypothesis: If the a change in bank resolution regimes (e.g. in the legal provisions
governing bank resolution) indeed represents a credible and e�ective improvement to bank resolution
technology, it will change the behavior of those �nancial institutions a�ected towards less risk-taking
and safer business models. We thus expect a decrease in risk measures for a�ected banks after the
change becomes e�ective.
Extended hypothesis I: The above e�ect might vary with the credibility and the political
will to truly resolve failed institutions. Both credibility and political will can be in�uenced and hence
proxied by exogenous (e.g. elections, overall state of the economy) or endogenous (e.g. characteristics
of the bank such as systemic importance that in�uence the discipline-liquidity trade-o�) variables. If
the application of the new regime is not credible due to bank-speci�c characteristics, we expect to �nd
a lower or even no e�ect on the respective banks' risk-taking.
Extended hypothesis II: Changes in bank regulation that reduce banks' charter value might
lead to gambling, particularly during the time after public announcement and before legal enactment.
If the political and legislative procedures around the introduction of changes in bank resolution regimes
provide opportunities for gambling, we expect to see an increase in risk measures for the a�ected banks
after announcement and before e�ective enactment of the change.
3 Identi�cation strategy - An application to changes in the U.S.
bank resolution regime
While the existing literature and the theoretical model of DeYoung et al. (2013) provide testable
implications of changes in resolution regimes, the actual empirical testing is challenging due to the
endogenous relation between bank behavior and resolution. In order to overcome these endogeneity
concerns and to test our hypotheses formulated above, we apply the theory of failed bank resolution
to a speci�c change in the U.S. bank resolution regime, the introduction of the Orderly Liquidation
Authority. We argue that the circumstances of the OLA introduction resemble a natural experiment
setup that can be exploited using a di�erence-in-di�erence model. This section describes the �t of this
speci�c resolution regime change and the identi�cation strategy by (1) discussing whether the OLA
indeed constitutes an improvement in resolution technology (i.e. whether it can indeed be taken as a
relevant treatment), (2) timing the introduction of the OLA (i.e. the treatment e�ect), (3) de�ning
di�erentially a�ected �nancial institutions (i.e. treatment and control group). Finally, we present
some initial evidence that supports our identi�cation setup and merits the more formal evaluation in
the sections to follow.
3.1 Identifying the treatment - Is the Orderly Liquidation Authority an
improvement in resolution technology?
When the �nancial crisis hit in 2008 (and surely before), U.S. bank resolution law su�ered from two
signi�cant shortcomings. We will argue that the Orderly Liquidation Authority represents a signi�cant
technological improvement on these two issues.
As a �rst issue, �nancial institutions in the U.S. were subject to two di�erent insolvency and
resolution regimes. One pillar of bank insolvency legislation was the Federal Deposit Insurance Act
(FDIA), that covered all insured depository institutions, particularly commercial banks, thrifts, and
savings banks holding a national or state charter. The FDIA stipulates a special resolution regime
for these institutions - an administrative insolvency procedure. The existence of this special bank
7
resolution regime stems from the conviction that banks are somewhat distinctive, particularly with
regard to insolvency. Marin and Vlahu (2011) provide a detailed analysis of the characteristics of
banks that advocate a special resolution regime - among the most important ones are (1) the inherent
instability of banking and the threat of runs, (2) particularly negative externalities of bank failures,
and (3) the potential for moral hazard due to deposit insurance schemes or implicit guarantees. While
the corporate insolvency law does not cover these aspects explicitly, the FDIA regime takes the special
role and functioning of �nancial institutions into account. It is designed to allow timely intervention
and resolution of insolvent banks while limiting moral hazard as well as potentially detrimental e�ects
to liquidity, sound banks, and the real economy. In order to achieve the goal of a least cost (and
least adverse e�ects) resolution, the special resolution regime deviates signi�cantly from the regular,
judicial insolvency procedure with regard to insolvency triggers and initiation conditions, resolution
instruments, �nancing, and possibilities for appeal and review (Bliss and Kaufman, 2006; Marin and
Vlahu, 2011). Under these provisions, the Federal Deposit Insurance Corporation (FDIC) has powers
to promptly intervene upon certain initiating conditions, such as critical undercapitalization, without
having to wait for the �ling of a default event or for court decision. In this case, the license of the
bank can be revoked by its primary regulator and the FDIC can be determined as the conservator or
receiver, ousting management and shareholders, taking over the bank, and ultimately preparing it for
purchase and assumption by another �nancial institution or for closure and liquidation. In order to
preserve liquidity, charter value, and operations of the bank, the FDIC typically intervenes overnight
or over the weekend and is able to pay o� all insured depositors - if need should be - from the Deposit
Insurance Fund previously collected from insured institutions (Bliss and Kaufman, 2006; DeYoung
et al., 2013).
While the FDIA covers insured depository institutions under national and state bank charters, the
FDIC did not have legal powers for intervention when it comes to the failure of bank holding companies,
�nancial holding companies, or other non-bank �nancial institutions. Instead, the default legal provi-
sions of corporate insolvency law, i.e. the insolvency procedures according to Chapter 7 and Chapter
11 of the U.S. Federal Bankruptcy Code, applied. These procedures typically protect the owners from
creditors, take long time periods for resolution during which funds for depositors and borrowers might
not be available, and require a restructuring plan as a precondition before making decisions on larger
asset sales (DeYoung et al., 2013). Since the �nancial holdings and non-bank �nancial institutions in
question - among them several of the institutions that have been identi�ed as systemically important
- exhibit similar characteristics to banks as described by Marin and Vlahu (2011), an application of
these corporate insolvency procedures might cause severe disruptions.9 While these institutions were
e�ectively exempted from the special bank resolution regime, the default corporate law was appar-
ently inappropriate to e�ciently resolve their insolvency. Hence, this was widely considered as a major
de�ciency in the resolution regime for �nancial holdings and non-bank �nancial �rms, which might
have even protected these institutions from actual failure by making bailout the only available choice
(FDIC, 2011; Marin and Vlahu, 2011).
Moreover, even if the FDIC had been legally empowered to apply its resolution procedure to non-
bank �nancial institutions, there would have been a �nancial limit as to which institutions it could
have e�ectively taken over: While the Deposit Insurance Fund amounted to a record high of USD 52.4
billion at the onset of the �nancial crisis, the deposits of Bank of America alone were about 10 times
larger than this (albeit not all insured). The sheer order of magnitude of this di�erence illustrates the
second signi�cant issue gripping the resolution technology available to U.S. regulators before 2010: Not
just incomprehensive legal provisions, but also the insu�cient �nancial endowment of the regulator
9In fact, several studies examine the inapplicability of corporate insolvency law to �nancial institutions, e.g. referingto one of the few bankruptcy cases of �nancial �rms: Lehman Brothers Holding Inc. (FDIC, 2011).
8
prevented an e�ective application of bank resolution and made bailout the regulator's preferred choice
in most cases for �nancial holdings and non-bank �nancial companies.10
Recognizing the need for alterations in bank resolution law and for stepping-up operational and
�nancial capabilities of the regulator, U.S. federal legislators passed the Orderly Liquidation Authority
as part of the wider �nancial sector reform package, the Dodd-Frank Act (DFA, Title II). The new
provisions stipulated by the OLA can be considered as an improvement to resolution technology in
several dimensions. First, it extends a special insolvency and resolution regime to �nancial institutions
previously uncovered by bank resolution law. More speci�cally, it stipulates that any �rm determined as
a covered �nancial company according to Sec. 201 and 203 of the DFA can be put into an administrative
insolvency and resolution procedure. E�ectively, this provision covers any �nancial institution in the
United States.11 The determination of a �nancial institution as a covered �nancial company is made by
the Secretary of the Treasury, following the vote of the FED board and FDIC board, and in consultation
with the President. It initiates the orderly liquidation procedure, with only limited judicial appeal ex
ante.12 Technically, this procedure is very similar to the existing FDIA regime, with the FDIC being
appointed as receiver of the �nancial company. Once under receivership, the FDIC is empowered to
close and liquidate the �rm, to pursue a purchase and assumption resolution, or to set up a bridge
�nancial institution. These resolution instruments also resemble the FDIA regime insofar as they cause
losses to shareholders and unsecured creditors, replace the management, and protect liquidity in a way
that is superior to regular insolvency law.
Second, Title II of the DFA sets up a new Orderly Liquidation Fund that also �nancially enables
the FDIC to act as the receiver and pursue the orderly liquidation of covered �nancial companies.
While the fund is set up in the Treasury, the FDIC is authorized to borrow from it for covering the
cost of orderly liquidation and administrative expenses. Moreover, the FDIC is empowered to charge
ex post risk-based assessments to �nancial companies13 in order to repay the Orderly Liquidation Fund
(DFA, Title II, Sec. 210).
Taken together, the Orderly Liquidation Authority can be interpreted as an improvement to reso-
lution technology (in the sense of DeYoung et al. (2013)) in at least two dimensions. First, we interpret
the OLA as an improvement in terms of legal authorities as it alleviates the previous limitation of the
FDIC to only place a certain group of �nancial institutions into a special bank resolution procedure.
Rather than focusing only on insured depository institutions, the special resolution regime is now
extended to other �nancial companies as well. Second, the establishment of the Orderly Liquidation
Fund signi�cantly improves the �nancial and operational capacity of the FDIC to e�ectively act as
receiver and liquidity guarantor. There is now less reason to prefer bailout over resolution when large
�nancial institutions fail - at least theoretically. These improvements might not establish an optimal
and ultimate resolution regime - rather, there is a broad discussion in the literature suggesting changes
10It should be noted that bailout was not preferred for a myriad of smaller banks that were covered by the FDIAand for which the Deposit Insurance Fund proved large enough: Between 2008 und 2010, the FDIC resolved the recordnumber of more than 300 banks.
11The determination as a covered �nancial company essentially requires three conditions to be ful�lled. Firstly, the �rmin question needs to be a �nancial company, i.e. a bank holding company, a non-bank �nancial company supervised bythe FED board, or any company predominantly engaged in �nancial activities. Secondly, it is not an insured depositoryinstitution covered by the FDIA regime. Finally, the determination is made provided the existence of all criteria outlinedin Sec. 203b, i.e. the �rm is in (danger of) default, the resolution according to otherwise applicable legal provisionswould have adverse consequences for �nancial stability, there is no viable private sector alternative, impact on creditorsand shareholders is appropriate, all convertible debt has been ordered to be converted, and the OLA is deemed e�ective(DFA, Title II, Sec 201, 203).
12In fact, the board of the determined covered �nancial company can ask the Secretary of the Treasury to petitionfor a formal authorization by the U.S. district court in the District of Columbia. This court can order the authorizationafter �nding that the determination as covered �nancial company is not arbitrary and capricious. If the court does notdecide within 24 hours, the authorization is automatically granted by the operation of law (DFA, Title II, Sec. 202).
13More speci�cally, Sec. 210 stipulates that the assessments are to be imposed on large non-bank �nancial institutions,precisely bank holding companies with consolidated assets exceeding USD 50 billion and non-bank �nancial companiessupervised by the FED board.
9
that might be even more appropriate (Bliss and Kaufman, 2011; Edwards, 2011; Fitzpatrick et al.,
2012; Scott et al., 2010; Scott and Taylor, 2012; Zaring, 2010). However, most of these commentators
(and the leading �nancial press quoted in the prelude of this paper) agree that the Orderly Liquidation
Authority at least represents a theoretical improvement to the pre-existing regime. In fact, DeYoung
et al. (2013) themselves describe it as a 'positive technological shock for U.S. bank regulators' and add
the prediction that (if e�ective) this will make resolution of insolvent �nancial institutions more likely
and hence reduce their incentives to choose high-risk business strategies.
Hence, we argue that the introduction of the OLA is indeed a signi�cant improvement to resolution
technology and will use it as the treatment whose e�ect we test going forward.
3.2 Timing the treatment - When did the treatment take place?
As with any legislative process, the introduction of the OLA stretched over a signi�cant timespan from
the generation of the idea to the bill being passed and signed into law by the President. The earliest
proposal for legislation regarding an Orderly Liquidation Authority was contained in the �nancial sector
reform package suggested by the Obama administration in June 2009 (Department of the Treasury,
2009). A revised proposal for the Orderly Liquidation Authority was announced as part of the reform
package that was later named the Dodd-Frank Act in December 2009. The major legislative process
took place over the following six months in the House of Representatives and the Senate. Finally, the
Dodd-Frank Act (and with it the OLA) was passed by the U.S. Congress in July 2010 and signed into
law by President Obama on July 21, with immediate e�ect. For our purposes, the �rst indication when
banks were confronted with the likely change of regulation planned by the Obama administration (June
2009) until the actual enactment of the legislation (July 2010) can be understood as the treatment
period.
Since our dataset is constructed from quarterly data, we de�ne all periods before and including the
second quarter of 2009 as pre-treatment periods and all periods after and including the third quarter
2010 as post-treatment periods.14
3.3 Identifying treatment and control group - Where �nancial institutions
di�erentially a�ected?
An important pillar of our identi�cation strategy is the di�erential e�ect of the OLA on �nancial
institutions that was already indicated above. While insured depository institutions were subject to
bank resolution law before, other �nancial institutions - speci�cally bank holding companies (BHCs)
and non-bank �nancial companies - were de facto not resolvable in an appropriate manner due to
the legal inapplicability of the FDIA and the economic inapplicability of corporate bankruptcy law.
Essentially, the introduction of the OLA only a�ected the latter group by exposing them to a credible
threat of resolution for the �rst time.
However, the actual situation is a bit less clear cut, as most holding companies own bank subsidiaries
that fall under the FDIA resolution authority.15 In some cases, the bank subsidiary even makes up
99% of the holding's assets, with the holding just being a legal mantle used for accounting, tax, and
other purposes. In order to not treat the constructs that have 99% of assets regulated by the FDIA
and those that only have 10% the same way, we propose an indicator that measures the share of
assets of a holding company not subject to the FDIA resolution regulation. In our view, this indicator
14Due to data availability and data quality we have to de�ne slightly di�erent pre- and post-treatment periods in theloan level dataset. Refer to the following section for more details.
15As indicated in the prelude, even Goldman Sachs Financial Holding owned subsidiaries (such as the Goldman SachsBank) that fall under the de�nition of an insured depository institution and were hence subject to resolution proceduresgoverned by the FDIA.
10
has the advantages that it captures the essence of our identi�cation idea and is simple to compute.
While we can also use the continuous indicator to build an interaction term, we will start with a pure
di�erence-in-di�erence setup by de�ning cuto�s that identify treatment and control group. We de�ne
all BHCs (and banks belonging to a BHC) that hold more than 30% non-FDIA-regulated assets as
particularly 'a�ected' by the regulatory change, i.e. as treatment group. On the other hand, we de�ne
all BHCs (and banks belonging to a BHC) that do not have any or less than 10% non-FDIA-regulated
assets as 'not a�ected', i.e. as control group. However, since these cuto�s are admittedly arbitrary, we
test several alternative cuto�s as well as the use of the continuous indicator in our robustness checks.
Taking the di�erential exposure to FDIA regulation as the criterion for distinguishing treatment
and control group enables us to employ a di�erence-in-di�erence setup to estimate the e�ect of OLA
on risk-taking. As our key identifying propositions, we have to assume that (1) treatment and control
group are developing in parallel (but not necessary at the same level) and (2) only the treatment
a�ected the treatment and control group di�erently (i.e. what we measuring is actually the treatment
e�ect and not something else). We argue that both is the case and present evidence for (1) in the
following sections. Regarding the di�erential treatment e�ect (2), we assume that most other changes
that took place at the same time as the introduction of the OLA concerned banks independently of their
share of assets under FDIA regulation. The �rst argument supporting this assumption is that among
several regulatory changes that took place at that time, the introduction of OLA is regarded the most
in�uential one (see, e.g., the quote from the Economist in the prelude). Secondly, other changes might
have been discussed or passed in the context of the Dodd-Frank Act, but many of them only became
e�ective at later dates.16 And, thirdly, even if other important changes (in regulation or other aspects
of the banking business) became e�ective at the same time, none of those did arguably a�ect banks
di�erentially depending on their share of FDIA regulated assets. Finally, one might rather argue that
BHCs with large unregulated share run a very di�erent business model and are hence (assuming that
this cannot be controlled for by covariates and �xed e�ects, which we will actually do) experiencing a
di�erential e�ect from other regulatory or �nancial market changes that took place at the same time.
This would, for example, be the case for holdings with large investment banking or trading units,
which were a particular target of the regulation that was passed at that time. Succumbing to this line
of reasoning, we resort to the bank level (besides using it as a robustness check) where these e�ects
should not be pronounced. Rather, the business models of insured depository banks (the ones that
are individual banks or belong to an a�ected group vs. the ones that belong to a non a�ected group)
should be far more comparable as the business models of a holding with large investment banking
departments and a holding where depository banking represents 99% of the assets.
Nevertheless: To the extent that parallel changes might have a�ected banks' risk-taking propor-
tionally to their non-FDIA-regulated share as well, we would also pick up their e�ect in our estimates.
While we are convinced not to �nd such e�ects outside the regulatory reform area, it could admittedly
be that, for example, regulatory attention to mostly non-FDIA-regulated institutions increased with
the introduction of the new resolution law. Hence, we should be aware that we are not just measur-
ing the e�ect of a mere change in the law, but in the whole resolution regime, including credibility,
capability (e.g. the Orderly Liquidation Fund), and attention of the regulator that this legal change
evoked.
3.4 Initial evidence - Does it really make a di�erence?
Is the OLA a technological improvement that is credible and e�ective? Is there enough political will
to use it? Does this new threat invoke a change in bank behavior, particularly for the most a�ected
16See, for example, the detailed overviews of implementation timelines and e�ective dates produced by Anand (2011);CCH Attorney-Editor (2010); DavisPolk (2010).
11
Figure 1: Change in z-score by non-FDIA-regulated asset share
institutions, i.e. those ones covered by a special resolution regime for the �rst time?
Figure 1 provides a �rst indication that the non-FDIA-regulated share could indeed be related
to changes in bank risk-taking after the introduction of the OLA: We plot the average di�erence in
overall bank risk between the pre- and post-treatment over ranges of non-FDIA-regulated share. As
a measure for bank risk, we use the average z-score, which is a composite measure approximating the
inverse of the default probability, i.e. higher z-scores stand for less overall bank risk.17 Although this is
only a very rough indication, it is interesting to note that higher ranges of non-FDIA-regulated shares
correspond to higher increases of the z-score, i.e. lower overall bank risk, after the introduction of the
Orderly Liquidation Authority.
Figure 2 and 3 provide an intuition of how a�ected (i.e. treatment) and non-a�ected (i.e. control)
banks' overall risk develops over a longer time and reacts to the introduction of the Orderly Liquidation
Authority. Again, we depict the average z-score of each group as a measure for overall bank risk -
this time taking the absolute values and evaluating them over time. Since the z-score incorporates the
standard deviation of returns, we have to compute it over a period stretching several quarters. We do
this for 8-quarter periods (Figure 2) and 4-quarter periods (Figure 3), both pre- and post-treatment,
excluding the treatment period as de�ned above (Q3 2009 - Q2 2010).
Admittedly, this is only a very crude evaluation that does not control for potentially omitted
variables and other sources of endogeneity beyond the bivariate di�erence-in-di�erence setup. But
several interesting patterns emerge from the two �gures. First, the di�erential behavior of a�ected
and non-a�ected banks around the treatment is evident: In both �gures, the a�ected banks experience
a much stronger increase in the z-score between the pre-treatment and the post-treatment period.
However, the key identifying assumption of di�erence-in-di�erence is that the two groups would exhibit
a parallel development in the absence of treatment. We can test this parallel trend assumption by
17Refer to the following section for a detailed description of the composition of the z-score.
12
Figure 2: Bank risk-taking before and after OLA
Figure 3: Bank risk-taking before and after OLA
13
including additional periods of data before and after the pre- and post-treatment period. Indeed, we
�nd a parallel trend before the treatment: In both graphs, a�ected and non-a�ected institutions develop
approximately in parallel in the absence of treatment. Figure 3 even allows us to add an additional
period after the post-treatment period, which again exhibits a parallel trend. It is interesting to observe
that a�ected banks consistently exhibit higher risk (lower z-score) before the treatment and reverse
this after the treatment. Taken together: In the absence of treatment, both a�ected and non-a�ected
banks seem to develop in parallel. It is only at the introduction of the OLA, that the treatment group
of a�ected banks experiences a materially di�erent behavior, i.e. a larger decrease in risk-taking, as
compared to the control group of non-a�ected banks. Consequently, these results are a �rst indication
that our main hypothesis might be correct. We test both the main hypothesis as well as the parallel
trend assumption in a more rigorous empirical framework below.
4 Model and dataset
4.1 Baseline model
For a more rigorous empirical testing, we construct a di�erence-in-di�erence model whose baseline
version is depicted in equation 1. The main dependent variable of the model is Riski,t, one of the risk
measures outlined below. The core explanatory variables are afterOLAt indicating before or after
treatment (i.e. improvement in resolution technology) and AFFECTEDi being a dummy variable
set to 1 for those institutions a�ected by the improvement in resolution technology and to 0 for the
control group (non-a�ected). Bank (γi) and time (δt) �xed e�ects are used to control for in�uences
constant either over time (e.g. time-invariant bank characteristics) or across banks (e.g. state of the
economy or the �nancial system in a speci�c quarter). The model is complemented by a set of control
variables (Xi,t) to control for additional covariates that might vary by treatment and control group
and in�uence bank behavior. If our main hypothesis holds true, we expect to see a decreasing e�ect of
the di�erence-in-di�erence term on risk, expressed in the direction and signi�cance of parameter β3.
Riski,t =α+ β1 ∗ afterOLAt + β2 ∗AFFECTEDi
+ β3 ∗ (afterOLAt ∗AFFECTEDi) (1)
+ γi + δt +Xi,t + εi,t
In order to ensure the robustness of our results, we test our hypotheses on di�erent levels and using
alternative empirical setups and datasets. First, we identify bank level data from quarterly call reports
that we merge with data from quarterly BHC reports in order to construct a dataset covering �nancial
data on bank- and BHC-level. This dataset enables us to compute and test bank level risk measures
as dependent variables in the above setup. Additionally, we de�ne several measures for business model
choices (e.g. regarding portfolio decisions or funding structure) that can be tested on the bank level.
Second, we investigate risk-taking decisions on the level of new mortgage loan business. Therefore, we
construct a loan level dataset using the Home Mortgage Disclosure Act (HMDA) Loan Application
Registry.
4.2 BHC and bank level dataset
We construct the bank level dataset based on two main sources. On the individual bank level, we
assemble data from the Consolidated Reports of Condition and Income (FFIEC031/041), commonly
known as call reports. These reports cover several hundred items of �nancial data, which any bank
with a state or national charter is required to �le on a quarterly basis with the FFIEC. We construct
14
a sample that contains the full set of banks and �nancial data for the period covering the �rst quarter
of 2005 till the second quarter of 2012. In addition, we assemble a second dataset on the Bank Holding
Company level. BHCs are required to �le quarterly �nancial reports on a consolidated and parent-only
level (FR Y-9C/LP/SP), which are available from the FED Chicago. As for the individual bank data,
we construct a sample that contains the full set of BHCs and selected �nancial data for the period
covering the �rst quarter of 2005 till the second quarter of 2012. In a third step, we obtain identi�ers
for the top-holders, i.e. the ultimate owner, of any individual bank from the FDIC's Statistics on
Depository Institutions (SDI), to match both the individual bank level and the BHC level datasets.
This matched dataset enables us to identify and compute all necessary variables as de�ned below.
Panel A of Table 1 provides summary statistics and sources of the data on the BHC level, while Panel
B focuses on the bank level data.
Dependent variables: Risk and business model measures In order to conduct a series of
robustness checks, we use several measures of risk-taking on the overall bank (or BHC) level. Our
primary measure is the z-score of each bank, which is de�ned as Z = (RoA + CAR)/σRoA, with
RoA being the mean return on assets, CAR the capital asset ratio, and σRoA the estimated standard
deviation of the return on assets. Mean and standard deviation of return on assets are computed
over 8-quarter periods (and additionally over 4-quarter periods for robustness tests). Very few banks
for which less than 3 datapoints in one of the periods are available for this computation are removed
from the sample. The z-score has been widely used in the empirical literature as a proxy for overall
bank risk (e.g. Boyd et al. (2010); Dam and Koetter (2012); Gropp et al. (2010); Laeven and Levine
(2009); Roy (1952)). Essentially, it measures the number of standard deviations by which a bank's
return on assets would have to fall below its mean in order to deplete the available equity. If we de�ne
default as losses exceeding equity, the z-score can be interpreted as a measure for distance to default
or the inverse of the default probability (Laeven and Levine, 2009; Roy, 1952). Hence, a higher z-score
indicates that a bank is more stable, i.e. is associated with less overall risk. We follow Laeven and
Levine (2009) in computing the natural logarithm of the z-score.18
In addition, we use the σRoA as an alternative risk measure that focuses exclusively on the volatility
of banks' return on assets. The return volatility has been used as a measure for overall bank risk in
several empirical studies before (e.g. Dam and Koetter (2012); Laeven and Levine (2009)). We
complement the z-score and σRoA with an alternative overall risk measure - average asset risk -
which is de�ned as RWA/assets, with RWA being the risk-weighted assets. This measure gives an
indication of average asset risk (albeit only in a pre-de�ned, regulatory sense) and has also been used
in the empirical literature (e.g. Berger et al. (2012); De Nicolò et al. (2010)). While the average asset
risk is a relatively simple measure and risk weights have been criticized as inadequate expression of
true risk, this measure o�ers the advantage to be computable on an individual quarterly level. In any
case, we use alternative risk measures as dependent variables to test the robustness of our results.
In order to test the impact of the regulatory change on the business model of banks, we also de�ne
a set of additional dependent variables that proxy for business model choices. With regard to portfolio
risk choices, we use some of the measures suggested by Duchin and Sosyura (2012). In detail, these
are the trading asset ratio (de�ned as ratio of assets held in trading accounts to total assets), the low
risk securities ratio (de�ned as the ratio of securities of U.S. government agencies and subdivisions to
total securities), and the high risk securities ratio (de�ned as the ratio of equity securities, asset-backed
securities, and trading accounts to total securities). Additionally, we use the CRECD loan ratio, which
is de�ned as the sum of commercial real estate loans (CRE) and construction and development loans
(CD) divided by total loans. This ratio is used as a proxy for the degree of complex and risky loans
18As the z-score is highly skewed, its natural logarithm is assumed to be approximately normally distributed.
15
on a bank's balance sheet and has been shown to be associated with risky business models more prone
to bank failure (e.g. DeYoung (2013)).
Beyond the asset side, we also take a measure from the liability side of banks' balance sheets into
account. More precisely, we test the e�ect on the deposit funding ratio, which is simply de�ned as
deposits divided by assets. This measure is intended to capture the riskiness of the funding structure
and the vulnerability to liquidity shocks.
Finally, we also de�ne a measure for risk in income structure. For this, we use the non-interest
income ratio, which we compute as average non-interest income divided by average total income.19
Non-interest income, particularly from non-core activities such as investment banking, venture capital
and trading activities, has been shown to be relatively volatile compared to interest income (DeYoung
and Roland, 2001) and to be associated with higher overall bank risk (Brunnermeier et al., 2012;
DeJonghe, 2010; Demirgüç-Kunt and Huizinga, 2010).
It should be noted that all of the dependent variables are calculated from accounting data, using
the call reports and BHC reports datasets. Despite their known shortcomings, we prefer accounting
data over market data as the latter would signi�cantly reduce our sample size, particularly for smaller
banks.
Explanatory variables and controls In accordance with the identi�cation strategy and the base-
line model outlined above, the treatment dummy AFFECTEDi, the treatment-period indicator
afterOLAt, and particularly the interaction between the two are de�ned as our main explanatory
variables. In order to identify the a�ected (i.e. treatment) group, we compute an indicator capturing
the non-FDIA-regulated share of total assets of a bank holding company. We do this by summing up
the total assets of all insured depository institutions (i.e. the ones that fall under the FDIA-regulation
and hence are subject to FDIC resolution authority) and scaling it by the total consolidated assets of
the BHC (including the non-bank, non-FDIA-regulated assets). For independent banks (i.e. insured
depository institutions that do not belong to a BHC), we set the non-FDIA-regulated share to 0. The
dummy indicating a�liation to the treatment group, AFFECTEDi, is set to 1 for all BHCs (and
banks belonging to a BHC in the bank level dataset) that hold more than 30% non-FDIA-regulated
assets, i.e. the group of BHCs and banks that is particularly a�ected. While the non-FDIA-regulated
share of assets varies between 0 and 100%, it is rather skewed towards the lower end, as most holding
companies own bank subsidiaries that fall under the FDIA resolution authority, some even exclusively.
A cuto� at 30%, however, delivers a su�ciently large treatment group. Moreover, a share of 30% is
arguably a signi�cant size of the total business of a bank, which will reasonably in�uence overall busi-
ness decisions and consequently a�ect institutions' behavior. At the lower end, we set AFFECTEDi
to 0 for all BHCs (and banks belonging to a BHC) that do not have any or less than 10% non-FDIA-
regulated assets. Admittedly, these cuto�s are highly arbitrary. Thus, we do not only use several
alternative cuto�s, but also an interaction with the continuous variable of non-FDIA-regulated share
of total assets to pursue additional robustness tests.
The second main explanatory variable, afterOLAt, is set to 1 for all periods between the third
quarter 2010 and the second quarter 2012. It is set to 0 for the eight quarters preceding the treatment,
i.e. from the third quarter 2007 to the second quarter 2009. To be able to formally test the parallel
trend assumption, we de�ne a second pre-pre-treatment period stretching over the eight quarters from
the third quarter 2005 to the second quarter 2007. As a robustness check, we use a second set of
afterOLAt and all variables referring to it, which de�nes afterOLAt over 4 quarters around the
treatment period.
In addition to the main explanatory variables, we control for a host of additional covariates that
19Note that we average over the 4- or 8-quarter periods de�ned above in order to balance single quarter e�ects.
16
might in�uence bank risk-taking and business model decisions, and that vary over banks and quarters
(i.e. that are not captured by the bank and time �xed e�ects in our model). In detail, these are total
assets as a proxy for bank size, capital ratio (de�ned as equity capital to total assets), return on assets
as a proxy for earnings capability, and liquidity ratio (de�ned as cash and balances at other depository
institutions to total assets). All of these variables are computed from the call report and BHC report
datasets. Furthermore, several recent analyses have shown that banks tend to increase risk when they
receive bailout assistance from the government, e.g. from the Capital Purchase Program (CPP) as
part of the Troubled Asset Relief Program (TARP) (Black and Hazelwood, 2012; Duchin and Sosyura,
2012). We follow these studies and add an indicator for the CPP status of a bank that is 1 if a bank is
a current recipient of CPP funds in a given quarter and 0 otherwise. Data for this indicator is obtained
from the U.S. Department of the Treasury CPP Transactions Report.
4.3 Loan level dataset
To test our hypotheses on risk-taking concerning new business operations - more speci�cally new
mortgage loan business - we use the HMDA Loan Application Registry as our loan level dataset.
HMDA requires most mortgage lenders to collect and report data on all mortgage loan applications on
an annual basis. According to Dell'Ariccia et al. (2012), the HMDA dataset comprises approximately
90% of all U.S. mortgage loan applications. The HMDA dataset is a comprehensive registry containing
loan information (e.g. loan purpose and loan amount), applicant information (e.g. race and gross
annual income), infomation on status of the loan application (e.g. sold, originated, denied, withdrawn)
including purchaser type or reasons for denial, and information on regional demographics. Moreover,
the dataset allows to distinguish beween supply and demand e�ects in the mortgage loan market.
The information whether the loan has been sold in the calender year of origination is very valuable
in our de�nition of actual risk-taking. Since approximately 60% of originated mortgage loans are
securitized (Loutskina and Strahan, 2009), we need to distinguish in our analyses between loans that
have been sold and loans that have been held on balance sheet at least for a certain time period,
because the former do not represent actual balance sheet risk-taking.20 A major disadvantage of the
HMDA dataset is that it does not provide more precise information on the time of loan application,
purchase, or origination than the calendar year.
We obtain all loan applications for the years 2009 to 2011 from the FFIEC.21 We remove three
sub-samples from the raw data: First, we exclude all loan applications that have been denied in the
pre-approval process, withdrawn or not accepted by the loan applicant or closed for incompleteness
to focus on those loans that have either been approved and originated or denied in the loan approval
process. Second, we drop all purchased loans from the sample to focus on true loan origination (and to
avoid double-counting of loans as the dataset does not allow for exact matching of sold and purchased
loans). Finally, we eliminate all loan applications with the purpose to re�nance an existing loan because
these loans usually have a di�erent pricing and underwriting structure than new home purchase or
home improvement loans (Avery et al., 2007).22 We supplement the HMDA dataset with data on the
regional housing price index obtained from the Federal Housing Finance Agency. We match the annual
20However, loans that remain on balance sheet do not necessarily represent balance sheet credit risk either, as lenderscan issue synthetic collateralized debt obligations on their loan portfolio to insulate credit risk while still retaining loanservicing. The HMDA dataset does not provide information on synthetic collateralized debt obligations. As a robustnesscheck we calculate the ratio of mortgage loans securitized but with servicing retained to total mortage loan portfoliofrom the bank level data and exclude all banks where this ratio is larger than 30%.
21This period is marked by a fall in house prices following the subprime mortgage crisis. We pay heed to account forthese adverse conditions as well as varying developments of the regional housing markets by adding regional housingmarket controls and regional �xed e�ects.
22Moreover, re�nancing loans could be biased due to 'evergreening' e�ects: Re�nancing loans can exhibit a higher riskpattern overall when they are intended to prolong non-performing home purchase loans that would be otherwise writteno�.
17
appreciation as well as the avarage annual level of the housing price index based on the Metropolitan
Statistical Area (MSA) in which the property is located.23 In a �nal step, we match this dataset with
the bank level dataset based on an individual and universal bank identi�er to identify treatment and
control group and to derive bank control variables.24 We use the bank level dataset since mortgage
loans are almost exclusively made through bank subsidiaries or individual banks.25 Panel C of Table
1 provides summary statistics for the resulting loan application sample.
Dependent variables We calculate the loan-to-income ratio (LIR) of each loan application as the
main risk measure in the loan level dataset. The LIR represents the loan applicant's ability to repay
the loan amount considering his gross annual income and indicates riskier loans by increasing loan-
to-income ratios. This measure is commonly used in the mortgage business to assess borrower risk,
e.g. it is a criterion for eligibility for loans insured by the Federal Housing Administration. According
to Dell'Ariccia et al. (2012), the measure is also used in lenders' loan decision processes. It usually
correlates strongly with other measures of individual loan risk: As shown by Rosen (2011), loans with
lower loan-to-income ratio tend to have better FICO scores.26 Therefore, we are con�dent that the
loan-to-income ratio is an appropriate risk measure in our loan sample. Since the distribution of the
loan-to-income ratio displays some distant outliers on the high end, we drop all loan observations with
loan-to-income ratio above the 99.5th percentile to avoid that our results are driven by those outliers.27
We do this trimming for the sample of loan applications as well as for the sample of originated loans,
so that the loan-to-income ratio ranges between 0 and 7.2 in our prepared sample. For the sample with
originated loans, we use the loan-to-income ratio as the dependent variable. For the sample of loan
applications, we exploit an approach similar to Duchin and Sosyura (2012). We simulate risk ranges
by dividing the full loan application sample into ranges with ∆ = 0.5LIR (0.0-0.5 being the safest and
>3.0 the riskiest loan-to-income range) and run our multivariate baseline model regression for each
range separately with the loan approval indicator as dependent variable. The loan approval indicator
is set to 1, if a loan application has been approved and originated and 0, if the loan application has
been denied. To rule out that our results are driven by loan demand than rather than by loan supply,
we calculate the natural log of the total number of loan applications received by a bank from each loan-
to-income range in each year and run our multivariate baseline model regression with this dependent
variable as in Duchin and Sosyura (2012).
Explanatory variables and controls We use the same explanatory variables in the loan level
dataset as described above. To identify treatment and control group in the loan level dataset, we use
the treatment dummy AFFECTEDi with the previously mentioned 10%/30% non-FDIA-regulated
asset share cuto�s. We also utilize the treatment dummy with di�erent cuto�s as a robustness check
and construct a continuous variable exploiting the share of non-FDIA-regulated assets. To distinguish
before and after treatment periods, we set the variable afterOLA to 1 for all loan applications in 2011
and to 0 for all loan applications in 2009.28
We control for several groups of additional covariates that might in�uence risk-taking in the new
mortgage loan business. First, we also use the set of bank control variables described above to ac-
23We use data for State Nonmetropolitan Areas when information on MSA is missing.24HMDA does not provide these indenti�ers for loans in 2009. We use identi�ers from 2010 and 2011 and match
lenders manually based on name and address when lenders are only present in the 2009 sub-sample.25We identify two lenders with BHC-status. For consistency reasons we exclude those observsations from our analyses.26FICO scores are provided by the Fair Isaac Corporation and measure a borrower's creditworthiness before obtaining
a mortgage loan.27We assume that these outliers mostly stem from misentries due to observed unrealistically high requested loan
amounts or very low annual incomes.28Since the calendar year is the only time designation in the HMDA dataset, we cannot match loans to particular
quarters. Due to current data availability from the FFIEC, we could not obtain loan applications for years prior to 2009.
18
count for bank size, capital adequacy, pro�tability, and liquidity. To capture further individual bank
characteristics, we exploit bank �xed e�ects. Second, we add dummy variables to control for certain
loan characteristics that indicate whether the loan has been sold and whether the loan is government-
guaranteed or -insured.29 Third, we control for demographic conditions by adding the log of total
population and the share of minority population for each U.S. Census Tract. Fourth, we take into
account economic conditions - especially the state of the housing markets - as these can signi�cantly
vary across U.S. regions. We control for the log of median family income as well as change and average
level of the house price index for each MSA. To capture further heterogeneity in demographic and eco-
nomic conditions that is not time-varying, we use regional �xed e�ects on a very detailed geographical
level, namely U.S. Census Tract (tract).
29Certain borrowers can receive loans that are insured by the Federal Housing Administration or guaranteed by theVeterans Administration, Farm Service Agency, or Rural Housing Services. Historically, these programs have allowedlower income U.S. borrowers to obtain mortgage loans that they could otherwise not a�ord.
19
Table 1: Summary statistics
Panel A: BHC sample
Variable group and name Source Mean SD Min Max N
Dependent variables (risk and business model)
Bank z-score BHC 4.57 (1.27) -2.76 11.96 46043
σ RoA BHC 19.09 (54.99) 0 2709 77613
Asset risk (RWA/assets) BHC 73.08 (11.98) 0 126.2 15395
Trading assets ratio BHC 0.33 (2.29) 0 42.75 14663
Low risk securities ratio BHC 0.21 (2.91) 0 100 15547
High risk securities ratio BHC 2.46 (9.37) 0 97.81 8797
CRECD loans ratio BHC 0.48 (1.64) 0 31.32 15642
Deposit funding ratio BHC 67.66 (13.41) 0 99.81 14663
Non-interest income ratio BHC 23.56 (14.29) 0.03 99.53 16679
Explanatory variables
BHC non-FDIA-regulated share BHC, SDI 12.23 (9) 0 100 46569
A�ected bank dummy (treatment) BHC, SDI 0.05 (0.22) 0 1 19467
After OLA dummy 0.49 (0.5) 0 1 86038
Additional bank- and quarter-varying control variables
Total assets (in USD mn) BHC 5040.52 (72044.57) 0 2358266 49112
Capital ratio BHC 10.04 (6.55) -57 100 47410
Earnings (RoA) BHC 0.1 (0.84) -41.95 81.82 47359
Liqudity ratio BHC 6.57 (6.61) 0.02 97.12 44375
CPP recipient bank-quarter TR 0.03 (0.18) 0 1 86038
Panel B: Bank sample
Variable group and name Source Mean SD Min Max N
Dependent variables (risk and business model)
Bank z-score SDI 4.44 (1.17) -9.46 8.83 126104
σ RoA SDI 25.58 (50.23) 0 2014.1 126427
Asset risk (RWA/assets) SDI 67.67 (14.72) 0 231.97 127022
Trading assets ratio SDI 0.07 (1.11) 0 77.17 126936
Low risk securities ratio SDI 71.36 (26.25) 0 100 123346
High risk securities ratio SDI 1.86 (9.17) 0 100 112917
CRECD loans ratio SDI 32.89 (20.88) 0 112.5 126209
Deposit funding ratio SDI 69.29 (11.45) 0 98.66 126785
Non-interest income ratio SDI 16.41 (12.65) 0 99.95 122973
Explanatory variables
BHC non-FDIA-regulated share BHC, SDI 7.68 (9.18) 0 100 89547
A�ected BHC dummy (treatment) BHC, SDI 0.03 (0.16) 0 1 56464
After OLA dummy 0.47 (0.5) 0 1 127170
Additional bank- and quarter-varying control variables
Total assets (in USD mn) SDI 1703319.62 (31321571.09) 66 1842568960 127170
Capital ratio SDI 11.72 (7.37) -13.52 100 126788
Earnings (RoA) SDI 0.11 (1.02) -28.38 93.5 126788
Liqudity ratio SDI 7.31 (7.93) 0 100 126936
CPP recipient bank-quarter TR 0.03 (0.17) 0 1 127170
Continued on next page
20
Table 1 � Continued from previous page
Panel C: Loan application sample
Variable group and name Source Mean SD Min Max N
Dependent variables
Loan-Income-Ratio (loan appl.) HMDA 2.04 (1.37) 0 7.22 4145701
Loan-Income-Ratio (orig. loans) HMDA 2.15 (1.29) 0 7.22 3106212
Loan-Income-Ratio (sold loans) HMDA 2.5 (1.13) 0.01 7.22 2021819
Loan-Income-Ratio (unsold loans) HMDA 1.5 (1.31) 0 7.22 1084393
Approval indicator HMDA 0.75 (0.43) 0 1 4329647
Explanatory variables
BHC non-regulated share (continuous) BHC, SDI 0.23 (0.21) 0 1 4089198
BHC non-regulated share (dummy) BHC, SDI 0.42 (0.49) 0 1 1876201
After OLA (2011/2009) 0.46 (0.5) 0 1 4329647
Additional bank control variables
Total assets (in USD mn) SDI 401968.92 (564608.08) 18.13 1788146.13 4329291
Capital ratio SDI 10.19 (2.6) -1.01 40.2 4329224
Earnings (RoA) SDI 0.12 (0.32) -6.08 2.36 4329224
Liqudity ratio SDI 5.69 (3.93) 0 77.74 4328745
CPP recipient bank TR 0.57 (0.49) 0 1 4329647
Additional loan, demographic and economic control variables
Government-guaranteed/-insured loan HMDA 0.3 (0.46) 0 1 4329647
Sold loan (orig. loans) HMDA 0.63 (0.48) 0 1 3242987
Total population in tract HMDA 5487.1 (2676.24) 1 36146 4280501
Minority population in tract HMDA 23.97 (25.29) 0.23 100 4280395
Median family income (in USD) HMDA 65698.53 (14446.18) 16100 111900 4280666
House price index level in MSA FHFA 183.56 (28.94) 110 338.02 4228877
House price index appreciation in MSA FHFA -3.67 (3.72) -19.49 9.21 4228877
Notes: This table reports variable names, sources, means, standard deviations, minimum and maximum values, and the
number of observations for which data is available in our sample. The sources are: FED Chicago BHC database (BHC),
Federal Housing Finance Agency (FHFA), Home Mortgage Disclosure Act Loan Application Registry (HMDA), FDIC SDI
database and call reports (SDI), U.S. Department of the Treasury (TR).
5 Results and robustness
This section presents and discusses our main results. We start from the e�ect of the improvement in
resolution technology on overall bank risk, and continue elaborating e�ects on bank business model
and loan decisions. These results are complemented by several extensions, e.g. testing the parallel
trend assumption using a placebo treatment event, tests for too-big-to-fail e�ects, as well as a search
for gambling behavior. Finally, we also discuss a set of robustness checks.
5.1 Overall bank risk-taking
In a very �rst step, we test the hypothesized e�ect of the OLA as an improvement in resolution
technology on overall bank risk, using a univariate version of our baseline model. Table 2 presents
the results of these univariate di�erence-in-di�erence comparisons, with Panel A focusing on a sample
containing individual bank data and Panel B comprising a sample of aggregated BHC data. The
21
treatment group includes all institutions that are particularly a�ected by the OLA, and is de�ned as
all banks (or BHCs in Panel B) belonging to a BHC with more than 30% of its assets not subject
to the FDIA resolution procedure. Conversely, the control group contains non-a�ected institutions,
i.e. all independent banks (that are hence fully subject to the FDIA resolution regime) and banks (or
BHCs) that are part of a holding with 10% or less non-FDIA-regulated assets.
For both the a�ected and non-a�ected institutions, we compute the means of the overall bank
risk measures before (Q3 2007 - Q2 2009) and after (Q3 2010 - Q2 2012) the introduction of the
Orderly Liquidation Authority. The resulting di�erences are tested for their statistical signi�cance and
displayed in columns (3) and (6). As a �rst result, it is interesting to note that all measures of overall
bank risk are signi�cantly decreasing across the board - for treatment and control groups on both bank
and BHC level - between the pre- and the post-treatment periods. This, however, is not necessarily
driven by the changes in regulation. Rather, it could be an overall trend towards less risk-taking that is
in�uenced by, e.g. macroeconomic trends.30 In order to test our hypothesis of a signi�cant di�erence
between treatment and control groups, we compute the univariate di�erence-in-di�erence results in
column (7). Interestingly, for both the z-score and σRoA measures, the treatment group experiences a
signi�cantly larger decline in overall risk between pre- and post-treatment as compared to the control
group. This �nding is fully in line with our main hypothesis. However, the picture for the asset risk
measure is less conclusive, as we do not �nd a signi�cant e�ect in the univariate di�erence-in-di�erence
estimates. Hence, these results may at most be interpreted as suggestive evidence - and we need to
proceed with more conclusive tests.
Since these results may also be driven by unobserved variables, we run multivariate di�erence-
in-di�erence estimations, adding two sets of �xed e�ects capturing both individual bank e�ects and
quarter e�ects as well as a set of time-variant control variables as outlined in the previous section.31
Panel A of Table 3 presents the results of these multivariate di�erence-in-di�erence estimations.32
These results show a highly signi�cant decline in overall risk between pre- and post-treatment for
a�ected banks as compared to non-a�ected banks. In particular, the coe�cient on the interaction
term afterOLAt ∗ AFFECTEDi is positive for the z-score (i.e. more stable), negative for σRoA
and asset risk (i.e. less volatile/risky), and statistically signi�cant at the 1 percent level for all risk
measures. These results hold both at the level of individual banks as well as on the level of BHCs
and strongly support our main hypothesis. Beyond statistical signi�cance, the results also suggest an
economically considerable impact: A�ected banks increase their z-score, for example, by more than
15%, which is about 3 times as much as non-a�ected banks.
In order to move beyond the arbitrary cuto�s de�ning the treatment and control groups, we also
estimate our model by replacing the treatment dummy with the actual share of assets not subject to
FDIA resolution. As before, we included bank and time �xed e�ects as well as time-variant controls
in our estimation. The results are displayed in Panel B of Table 3 and are very much in line with
our dummy results in Panel A. Again, the coe�cient on the interaction term indicates a signi�cant
increase in overall bank stability and a signi�cant decrease in overall bank risk. We also estimated
alternative cuto�s (e.g. 50 vs. 10 percent non-FDIA-regulated share of business) as robustness tests,
which are not reported but consistent with our main hypothesis.
The analyses presented so far have shown a signi�cant di�erence-in-di�erence e�ect, indicating that
risk-taking decreases with the degree to which a bank is a�ected by the improvement of resolution tech-
nologies. However, the validity of the di�erence-in-di�erence approach also relies upon the identifying
30One could, for example, argue that the outbreak of the �nancial crisis in 2008 increased volatility and that marketscalmed down after 2010, which causes the e�ect we �nd.
31Note that for brevity of the tables, we do not report the regression coe�cients on all of these control variables (whichare generally in line with expectations and previous empirical �ndings).
32Note that the level e�ect on the afterOLAt dummy drops as it is captured by the time �xed e�ects.
22
Table 2: Bank risk-taking: Univariate Di�erence-in-Di�erence analyses
Panel A: Bank level
(1) (2) (3)=(2)-(1) (4) (5) (6)=(5)-(4) (7)=(3)-(6)A�ected banks Non-a�ected banksBeforeOLA
AfterOLA Dif
BeforeOLA
AfterOLA Dif Dif-in-Dif
Dep. variable
Z-score 4.086 4.741 0.655*** 4.270 4.440 0.170*** 0.485***(0.0608) (0.0108) (0.0668)
σ RoA 0.521 0.234 -0.287*** 0.321 0.252 -0.0697*** -0.218***(0.0349) (0.00503) (0.0312)
Asset risk 0.694 0.631 -0.0618*** 0.681 0.630 -0.0517*** -0.0101(0.0014) (0.00132) (0.00822)
Panel B: BHC-level
(1) (2) (3)=(2)-(1) (4) (5) (6)=(5)-(4) (7)=(3)-(6)A�ected banks Non-a�ected banksBeforeOLA
AfterOLA Dif
BeforeOLA
AfterOLA Dif Dif-in-Dif
Dep. variable
Z-score 4.051 4.554 0.503*** 4.17 4.37 0.196*** 0.307***(0.0896) (0.0202) (0.0986)
σ RoA 1.119 0.409 -0.71*** 0.214 0.193 -0.0212*** -0.689***(0.196) (0.00477) (0.0475)
Asset risk 0.697 0.632 -0.0644*** 0.762 0.682 -0.0801*** 0.0157(0.0159) (0.00292) (0.0109)
Notes: This table presents univariate di�erence-in-di�erence estimates. Panel A reports the results for the bank sample,Panel B for the bank holding company (BHC) sample. Banks (or BHCs) are classi�ed into two groups. The treatmentgroup comprises a�ected banks (BHCs) that are part of a BHC with more than 30% of non-FDIA-regulated assets. Thecontrol group comprises non-a�ected banks (BHCs) that are independent or part of a BHC with less than 10% of non-FDIA-regulated assets. Treatment is de�ned as the introduction of the Orderly Liquidation Authority (OLA). Several measuresof overall bank risk are taken as dependent variables: z-score (de�ned as mean return on assets plus capital ratio dividedby the standard deviation of return on assets), σ RoA (de�ned as standard deviation of return on assets), and asset risk(de�ned as risk-weighted assets divided by total assets). Di�erence-in-di�erence estimates are displayed in column (7).
Heteroskedasticity consistent standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
23
Table 3: Bank risk-taking: Multivariate Di�erence-in-Di�erence analyses
Panel A: Dummy variable (treatment and control group de�nition)
(1) (2) (3) (4) (5) (6)Level Bank level BHC-levelDep. variable Z-score σ RoA Asset risk Z-score σ RoA Asset riskA�ected bank 0.131** 0.0459 0.0142
(0.0559) (0.0285) (0.00903)A�ected BHC -0.991*** -0.0649 -0.195
(0.253) (0.148) (0.141)A�ected bank x af-ter OLA 0.476*** -0.181*** -0.0220***
(0.0410) (0.0277) (0.00536)A�ected BHC x af-ter OLA 0.545*** -0.504*** -0.0131**
(0.0730) (0.153) (0.00645)
Constant YES YES YES YES YES YESControls YES YES YES YES YES YES
Bank FE YES YES YES YES YES YESTime FE YES YES YES YES YES YES
Observations 55,811 55,894 56,140 17,726 17,995 5,560R-squared 0.813 0.810 0.889 0.858 0.717 0.894
Panel B: Continuous variable (unregulated share in %)
(1) (2) (3) (4) (5) (6)Level Bank level BHC-levelDep. variable Z-score σ RoA Asset risk Z-score σ RoA Asset riskUnregulated share(parent BHC-level) 0.390*** 0.0151 0.0675***
(0.0673) (0.0277) (0.00948)Unregulated share(BHC-level) -0.869*** 0.162 -0.110
(0.244) (0.202) (0.0775)Unregulated sharex after OLA 0.772*** -0.133*** -0.0635***
(0.0537) (0.0276) (0.00690)Unregulated sharex after OLA 1.766*** -1.316*** -0.0338*
(0.155) (0.391) (0.0199)
Constant YES YES YES YES YES YESControls YES YES YES YES YES YES
Bank FE YES YES YES YES YES YESTime FE YES YES YES YES YES YES
Observations 88,710 88,795 89,194 43,050 43,338 14,221R-squared 0.786 0.797 0.885 0.809 0.743 0.877
Notes: This table presents multivariate di�erence-in-di�erence estimates of the e�ect that the introduction of theOrderly Liquidation Authority had on overall bank risk. Panel A reports the results for the di�erence-in-di�erenceestimation, Panel B for the estimation using a continous explanatory variable interaction. A�ected bank (BHC)takes a value of 1 if the bank (BHC) is part of a BHC with more than 30% of non-FDIA-regulated assets and avalue of 0 if the bank (BHC) is independent or part of a BHC with less than 10% of non-FDIA-regulated assets.Unregulated share is de�ned as the share of non-FDIA-regulated assets. After OLA is 1 for the quarters Q3 2010 -Q2 2012 and 0 for the quarters Q3 2007 - Q2 2009. Several measures of overall bank risk are taken as dependentvariables: z-score (de�ned as mean return on assets plus capital ratio divided by the standard deviation of returnon assets), σ RoA (de�ned as standard deviation of return on assets), and asset risk (de�ned as risk-weightedassets divided by total assets). Control variables comprise the natural logarithm of total bank assets, capital ratio,pro�tability, liquidity ratio, and an indicator variable that takes the value of 1 if the bank was a recipient of theTARP CPP program in a respective quarter (and 0 otherwise). All models include bank and time �xed e�ects.
Heteroskedasticity consistent standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
24
assumption of a parallel trend between treatment and control group in the absence of treatment. While
we presented some suggestive evidence underlining this assumption in the previous section, we now
apply a more rigorous approach in testing it. In order to do so, we extend our dataset to cover another
8-quarter period stretching from Q3 2005 to Q2 2007, which we de�ne as a pre-placebo period. We now
test the e�ect of a placebo treatment between the pre-placebo period and the pre-treatment period,
using essentially the same model as in the analyses above. If the parallel trend assumption holds,
we expect not to �nd a signi�cant di�erence-in-di�erence e�ect between the a�ected and non-a�ected
banks or BHCs across both periods. The results of this placebo test are displayed in Table 4. Indeed,
there is no signi�cant di�erence-in-di�erence e�ect to be found for the z-score and asset risk measures,
neither in the bank nor in the BHC panel. While the coe�cient on the interaction is also insigni�cant
with σRoA as dependent variable in the BHC sample, return volatility seems to increase for individual
banks belonging to an a�ected BHC after the placebo treatment. A potential explanation why it is
the return volatility (and just the return volatility) that is signi�cantly higher for a�ected banks could
be o�ered by the rational behavior that we would presume for these banks: As there is a lower threat
of resolution for these banks before the enactment of the OLA, they had incentives to take on higher
risks during the pre-placebo period (and before). When the �nancial crisis hits (which coincides with
the placebo treatment), this additional risk materializes in an overproportional increase of volatility.
Admittedly, this is only a vague explanation and further research is warranted to investigate this e�ect.
Apart from this one reaction of σRoA, however, the presented evidence is mostly consistent with the
parallel trend assumption.
Furthermore, it is interesting to note that the level e�ects for a�ected BHCs seem to con�rm the
presumption of higher overall risk of this group previous to the introduction of the OLA. This is
consistent with our hypothesis that holdings with high unregulated shares are less subject to FDIA
resolution and hence enjoy more of an implicit bailout guarantee - previous to OLA. The e�ect does not
occur for individual banks, presumably as these were already subject to FDIA resolution, even if they
were part of a BHC (implying that BHC risk-taking was largely done through the non-FDIA-regulated
parts). When the resolution threat becomes realistic for banks and BHCs alike (even if they hold high
previously non-FDIA-regulated shares), the di�erence in risk-taking and business model decisions is
hypothesized to occur both in a�ected banks and a�ected BHCs - which is remarkably consistent with
the results in previous and following tables.
Taken together, the results presented so far con�rm our main hypothesis: Banks or BHCs that were
largely not subject to the FDIA resolution regime before are particularly a�ected by the introduction
of the OLA and decrease their overall risk accordingly. We now want to go beyond the measures of
overall bank risk and analyze in more detail how banks change their behavior with regard to business
model and investment choices as well as new loan origination.
5.2 Bank business model choices and loan origination
As outlined above, we de�ne and compute several indicators for bank business model and investment
choices, that have been suggested in the literature (Brunnermeier et al., 2012; DeJonghe, 2010; DeY-
oung, 2013; Duchin and Sosyura, 2012). We test the di�erence-in-di�erence e�ect by using these indi-
cators as dependent variables in our multivariate baseline model, including �xed e�ects and additional
controls. Since data for these measures is in large parts only available at the bank level (particularly
for the loan data), we carry out all of our tests for the bank dataset. Table 5 presents the results,
which are all consistent with the hypothesized decrease in risky activities and investment choices for
the a�ected banks after the introduction of the OLA. We start with the e�ect on the trading assets ra-
tio (column (1)). In line with the expectation that a�ected banks decrease risky and volatile activities
(such as proprietary trading), we �nd a negative and signi�cant coe�cient on the interaction term. A
25
Table 4: Bank risk-taking: Multivariate Di�erence-in-Di�erence analyses with placebo test
(1) (2) (3) (4) (5) (6)Level Bank level BHC-levelDep. variable Z-score σ RoA Asset risk Z-score σ RoA Asset riskA�ected bank 0.160** -0.0706 -0.00704
(0.0639) (0.0468) (0.00900)A�ected BHC -1.084*** 0.382** 0.0586**
(0.242) (0.169) (0.0237)A�ected bank x af-ter placebo -0.0177 0.106*** 0.00590
(0.0367) (0.0214) (0.00362)A�ected BHC x af-ter placebo 0.0699 0.172 0.000800
(0.0804) (0.131) (0.00473)
Constant YES YES YES YES YES YESControls YES YES YES YES YES YES
Bank FE YES YES YES YES YES YESTime FE YES YES YES YES YES YES
Observations 62,757 62,792 63,122 20,017 20,075 7,740R-squared 0.755 0.819 0.901 0.787 0.774 0.933
Notes: This table presents multivariate di�erence-in-di�erence estimates for a placebo treatment. A�ected bank(BHC) takes a value of 1 if the bank (BHC) is part of a BHC with more than 30% of non-FDIA-regulated assets anda value of 0 if the bank (BHC) is independent or part of a BHC with less than 10% of non-FDIA-regulated assets.After placebo is 1 for the quarters Q3 2007 - Q2 2009 and 0 for the quarters Q3 2005 - Q2 2007. Several measuresof overall bank risk are taken as dependent variables: z-score (de�ned as mean return on assets plus capital ratiodivided by the standard deviation of return on assets), σ RoA (de�ned as standard deviation of return on assets),and asset risk (de�ned as risk-weighted assets divided by total assets). Control variables comprise the naturallogarithm of total bank assets, capital ratio, pro�tability, liquidity ratio, and an indicator variable that takes thevalue of 1 if the bank was a recipient of the TARP CPP program in a respective quarter (and 0 otherwise). Allmodels include bank and time �xed e�ects.
Heteroskedasticity consistent standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
26
Table 5: Bank business model and investment choices: Multivariate Di�erence-in-Di�erence analyses
(1) (2) (3) (4) (5) (6)Level Bank level
Dep. variable
Tradingassetsratio
Low risksecuritiesratio
High risksecuritiesratio
CRECDloan ratio
Depositfundingratio NII ratio
A�ected bank 0.00101 -0.0171 0.0404*** -0.00354 -0.0109 -0.000635(0.00318) (0.0225) (0.0151) (0.00862) (0.00720) (0.00647)
A�ected bank x af-ter OLA -0.00605*** 0.0584*** -0.0377*** -0.0108*** 0.0307*** -0.00927**
(0.00136) (0.0118) (0.00926) (0.00312) (0.00610) (0.00447)
Constant YES YES YES YES YES YESControls YES YES YES YES YES YES
Bank FE YES YES YES YES YES YESTime FE YES YES YES YES YES YES
Observations 56,140 54,000 44,050 55,384 56,137 53,737R-squared 0.776 0.778 0.784 0.961 0.907 0.921
Notes: This table presents multivariate di�erence-in-di�erence estimates of the e�ect that the introduction of theOrderly Liquidation Authority had on bank business model and investment decisions. A�ected bank takes a valueof 1 if the bank is part of a BHC with more than 30% of non-FDIA-regulated assets and a value of 0 if the bank isindependent or part of a BHC with less than 10% of non-FDIA-regulated assets. After OLA is 1 for the quartersQ3 2010 - Q2 2012 and 0 for the quarters Q3 2007 - Q2 2009. Several measures of bank business model andinvestment decisions are taken as dependent variables: trading asset ratio (de�ned as ratio of assets held in tradingaccounts to total assets), low risk securities ratio (de�ned as the ratio of securities of U.S. government agenciesand subdivisions to total investment securities), high risk securities ratio (de�ned as the ratio of equity securities,asset-backed securities, and trading accounts to total investment securities), CRECD loan ratio (de�ned as the sumof commercial real estate loans and construction and development loans, divided by total loans), deposit fundingratio (de�ned as deposits divided by assets), and non-interest income ratio (de�ned as average interest incomedivided by average total income). Control variables comprise the natural logarithm of total bank assets, capitalratio, pro�tability, liquidity ratio, and an indicator variable that takes the value of 1 if the bank was a recipient ofthe TARP CPP program in a respective quarter (and 0 otherwise). All models include bank and time �xed e�ects.
Heteroskedasticity consistent standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
similar result holds for the e�ect on the low and high risk securities ratios, presented in columns (2)
and (3): While a�ected banks seem to decrease investments in risky securities, they appear to increase
their exposure towards low risk securities classes. This shift in the securities portfolios is consistent
with the expectation that a�ected banks will rush for safer investments and business models after the
introduction of the OLA. In a similar vein, we would expect the treatment group of banks to decrease
their exposure towards highly complex and risky loans (such as the CRECD loans) relatively to their
total loan portfolio. The negative and signi�cant coe�cient on the di�erence-in-di�erence term in
column (4) suggests that we cannot reject this hypothesis.
Turning to the liability side of the bank business model, we would expect a�ected banks to opt
for sources of funding that are considered more stable and carry less interest rate risk. If the deposit
funding ratio correctly proxies for this, we �nd our expectation con�rmed by a positive and signi�cant
coe�cient on the interaction term. Finally, we look at the e�ect on the sources of income of the
bank. The negative coe�cient on the interaction term in column (6) suggests that a�ected banks
decrease their non-interest income relative to interest income stronger than the control group after the
introduction of the OLA. If non-interest income is indeed more volatile and associated with overall
(systemic) risk as claimed in the literature, the results found in column (6) are consistent with our
main hypothesis.
27
The data and evidence presented so far largely draws upon aggregated accounting data. In order
to complement this with actual risk-taking in business operations on banks' micro-level, we extend our
analysis to the mortgage loan business. We use our multivariate baseline model to test the di�erence-
in-di�erence e�ect on risk-taking in newly originated mortgage loans. Table 6 presents the results
exploiting the loan-to-income ratio as the risk measure. Column 1 displays an analysis on the whole
sample of newly originated loans, yielding a negative and signi�cant coe�cient on the interaction term
that con�rms our main hypothesis. In a second step, we split this sample into loans that have been
sold in the same calendar year (column (2)) and loans that have not been sold in the same calendar
year (column (3)). We assume that loans in the latter sample have been held on balance sheet at least
for a certain time period, so that they measure risk-taking more accurately. We �nd that a�ected
banks signi�cantly decrease loan-to-income ratios of new loans after the introduction of OLA for both
sold and unsold loans.
One further caveat could be loans that remain on the balance sheet for servicing but are de facto
securitizied (e.g. through synthetic collateralized debt obligations) and hence do not necessarily rep-
resent risk-taking. Since the HMDA dataset does not provide information on synthetic collateralized
debt obligations, we calculate the ratio of mortgage loans securitized but with servicing retained to
total mortgage loan portfolio from the bank level dataset and exclude all banks where this ratio of
synthetic loans is larger than 30%. We rerun our multivariate baseline model and �nd that a�ected
banks with a low share of synthetic loans in fact reduce the risk of new loans that remain on their
balance sheet after the introduction of OLA, while this e�ect is not signi�cant for sold loans (see Panel
B of Table 6).
It could be possible that our results on the sample of originated loans stem from loan demand
rather than loan supply e�ects, i.e. only high quality borrowers demand loans from a�ected banks
after the introduction of OLA. To account for potential loan demand e�ects, we include rejected loan
applications, split the loan application sample into di�erent risk ranges based on the loan-to-income
ratio, and test our main hypothesis using the application approval indicator as dependent variable. The
results for the analysis on the approval rate of loan applications are shown in Panel A of Table 7. We
�nd that the probability of loan approval by a�ected banks decreases after the introduction of OLA.
However, this decrease is not signi�cant for the safest risk range with loan-to-income ratio below 0.5,
while it is signi�cant for all remaining risk ranges. Interestingly, we �nd that in the pre-OLA period
the approval rate is higher for a�ected banks than for non-a�ected banks. For non-a�ected banks,
the approval rate also declines in the period after introduction of OLA, however only signi�cantly
in the safest loan-to-income ranges. Additionally, we test for systematic di�erences in loan demand
across risk ranges by employing the total number of loan applications per bank, year, and risk range
as dependent variable and �nd that the loan demand at a�ected banks did not signi�cantly decrease
after introduction of OLA (see Panel B of Table 7).
We bring forward evidence that after the introduction of the resolution threat, a�ected banks
decrease risk-taking in new loan business by approving less loans from higher risk ranges and can
exclude that our results are driven by loan demand e�ects. In sum, the presented results are consistent
with the interpretation that a�ected banks decrease their overall risk-taking after the introduction of
the Orderly Liquidation Authority and do so by shifting their investments, business models, and loan
decisions towards more prudent behavior.
28
Table 6: Risk taking in new mortgage loan business: Multivariate Di�erence-in-Di�erence analyses
Panel A: Newly originated loans from all banks in sample(1) (2) (3)
Level Loan levelSample All originated loans Sold loans Unsold loans
Dep. variable Loan-to-income ratioA�ected bank -0.685*** -0.170 -0.701***
(0.0767) (0.135) (0.0984)After OLA 0.00146 -0.0581*** 0.0458***
(0.00367) (0.00506) (0.00554)A�ected bank xafter OLA -0.0691*** -0.0352*** -0.0459***
(0.00477) (0.00603) (0.00918)
Constant YES YES YESBank controls YES YES YESLoan controls YES YES YESDemogr. controls YES YES YESEconomic controls YES YES YESBank FE YES YES YESTract FE YES YES YES
Observations 1,366,242 913,178 453,064R-squared 0.324 0.219 0.367
Panel B: Newly originated loans from banks with share of synthetic loans <30%(1) (2) (3)
Level Loan levelSample All originated loans Sold loans Unsold loans
Dep. variable Loan-to-income ratioA�ected bank -0.698*** -0.194 -0.747***
(0.0824) (0.136) (0.110)After OLA -0.0193*** -0.0624*** 0.00752
(0.00514) (0.00732) (0.00769)A�ected bank xafter OLA -0.0470*** -0.0192 -0.0406***
(0.00817) (0.0118) (0.0128)
Constant YES YES YESBank controls YES YES YESLoan controls YES YES YESDemogr. controls YES YES YESEconomic controls YES YES YESBank FE YES YES YESTract FE YES YES YES
Observations 830,560 532,525 298,035R-squared 0.350 0.229 0.387
Notes: This table presents multivariate di�erence-in-di�erence estimates of the e�ect that the introductionof the Orderly Liquidation Authority had on risk-taking in new originated mortgage loans. Panel A reportsthe results for the sample with all banks, Panel B restricts the sample to banks where the ratio of mortgageloans securitized but with servicing retained to total mortgage loan portfolio is less than 30%. Sold loansare originated loans that were sold in calendar year of origination; unsold loans are originated loans thatwere not sold in calendar year of origination. A�ected bank takes a value of 1 if the bank is part of a BHCwith more than 30% of non-FDIA-regulated assets and a value of 0 if the bank is independent or part ofa BHC with less than 10% of non-FDIA-regulated assets. After OLA is 1 for all loans originated in 2011and 0 for all loans originated in 2009. The dependent variable to measure risk-taking in new loans is theloan-to-income ratio. Bank control variables comprise the natural logarithm of total bank assets, capitalratio, pro�tability, and liquidity ratio. Loan control variables comprise two indicator variables: sold loanis equal to 1 if the loan has been sold (all originated loans sample) and guaranteed/insured loan is equalto 1 if the loan is guaranteed or insured by the government. Demographic control variables comprise thenatural logarithm of total population in tract and share of minority population in tract. Economic controlscomprise the natural logarithm of median family income in tract, appreciation and level of regional houseprice index. All models include bank and regional (tract) �xed e�ects.
Heteroskedasticity consistent standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
29
Table 7: Approval of mortgage loan applications and loan demand along risk ranges: MultivariateDi�erence-in-Di�erence analyses
Panel A: Approval rate of loan applications(1) (2) (3) (4) (5) (6) (7) (8)
Level Loan levelLoan applications within loan-to-income ratio range
Sample All appl. 0.0-0.5 0.5-1.0 1.0-1.5 1.5-2.0 2.0-2.5 2.5-3.0 >3.0Dep. variable Application approval indicator
A�ected bank 0.102*** 0.0270 0.0901 0.157** -0.0146 0.147* 0.172** 0.155*(0.0221) (0.0532) (0.0647) (0.0640) (0.0614) (0.0872) (0.0819) (0.0936)
After OLA -0.0043*** -0.0233*** -0.0117*** -0.00251 0.00423 -0.00275 -0.00112 0.00294(0.00103) (0.00345) (0.00358) (0.00317) (0.00266) (0.00254) (0.00272) (0.00218)
A�ected bank xafter OLA -0.0465*** -0.00640 -0.0167*** -0.0529*** -0.0630*** -0.0599*** -0.0540*** -0.0563***
(0.00127) (0.00491) (0.00481) (0.00406) (0.00336) (0.00319) (0.00339) (0.00253)
Constant YES YES YES YES YES YES YES YESBank controls YES YES YES YES YES YES YES YESLoan controls YES YES YES YES YES YES YES YESDemogr. controls YES YES YES YES YES YES YES YESEcon. controls YES YES YES YES YES YES YES YES
Bank FE YES YES YES YES YES YES YES YESTract FE YES YES YES YES YES YES YES YES
Observations 1,839,672 193,601 164,310 189,605 242,163 257,310 234,283 491,291R-squared 0.443 0.425 0.446 0.469 0.493 0.514 0.539 0.581
Panel B: Total number of loan applications(1) (2) (3) (4) (5) (6) (7) (8)
Level Loan levelLoan applications within loan-to-income ratio range
Sample All appl. 0.0-0.5 0.5-1.0 1.0-1.5 1.5-2.0 2.0-2.5 2.5-3.0 >3.0Dep. variable Log of total number of loan applications per bank, year, and range
A�ected bank -0.196 0.605 -0.216 -0.420* -0.230 0.101 -0.825*** -0.814**(0.180) (0.410) (0.278) (0.242) (0.299) (0.313) (0.242) (0.341)
After OLA -0.171*** -0.222*** -0.166*** -0.119*** -0.214*** -0.188*** -0.237*** -0.305***(0.0153) (0.0269) (0.0238) (0.0247) (0.0256) (0.0253) (0.0272) (0.0297)
A�ected bank xafter OLA -0.127 -0.229 -0.211 -0.198 -0.119 -0.109 -0.185 -0.0855
(0.122) (0.166) (0.133) (0.149) (0.178) (0.214) (0.238) (0.202)
Constant YES YES YES YES YES YES YES YESBank controls YES YES YES YES YES YES YES YES
Bank FE YES YES YES YES YES YES YES YES
Observations 33,762 4,510 4,492 4,338 4,225 4,060 3,791 4,261R-squared 0.015 0.085 0.078 0.072 0.097 0.104 0.108 0.157
Notes: This table presents multivariate di�erence-in-di�erence estimates of the e�ect that the introduction of the OrderlyLiquidation Authority had on approval rate of mortgage loan applications and loan demand along risk ranges. Column (1)shows the full sample of loan applications, columns (2)-(8) contain the sub-samples of loan applications based on loan-to-incomeratio ranges. The dependent variable in Panel A is the application approval indicator which equals 1 when loan applicationsucceeded in loan origination (and 0 when the application was denied). Panel B employs the natural logarithm of total numberof loan applications per bank, year, and risk range as dependent variable. A�ected bank takes a value of 1 if the bank is partof a BHC with more than 30% of non-FDIA-regulated assets and a value of 0 if the bank is independent or part of a BHC withless than 10% of non-FDIA-regulated assets. After OLA is 1 for all loan applications in 2011 and 0 for all loan applications in2009. Bank control variables comprise the natural logarithm of total bank assets, capital ratio, pro�tability, and liquidity ratio.Loan control variables comprise two indicator variables: sold loan is equal to 1 if the loan has been sold and guaranteed/insuredloan is equal to 1 if the loan is guaranteed or insured by the government. Demographic control variables comprise the naturallogarithm of total population in tract and share of minority population in tract. Economic controls comprise the naturallogarithm of median family income in tract, appreciation and level of regional house price index. Models in Panel A includebank and regional (tract) �xed e�ects; models in Panel B include bank �xed e�ects.
Heteroskedasticity consistent standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
30
5.3 Extensions and robustness
Is the OLA a credible threat for all banks or are there 'too-big-to-not-rescue' institutions?
So far, we have tested our main hypothesis and found that the a�ected banks indeed reduced their risk-
taking after the introduction of the OLA relative to non-a�ected banks. However, we also postulated
in the beginning that this e�ect might vary with the credibility and e�ectiveness, and in particular
with the political will to apply the new improvement in regulatory technology. Formulated in the
context of the model by DeYoung et al. (2013): When the political will or preference for discipline is
low or the liquidity trade-o� is high, we expect to �nd a lower or even no e�ect of the introduction
of the OLA on the behavior of a�ected banks. If �nancial institutions do not think that the OLA
represents a credible threat, they will not change their behavior as a response to it.
One straightforward - and admittedly simple - way of testing this prediction is the 'too-big-to-not-
rescue' e�ect. Essentially, we take systemic importance or sheer size of a bank as a proxy for high
liquidity trade-o�.33 Winding down such an institution might produce high liquidity costs, making
discipline less favored by regulators, which ultimately results in low credibility of the threat of resolution
- even after the introduction of the OLA. Such institutions are 'too-big-to-not-rescue'. Hence, we test
whether extraordinarily large institutions are less (or not at all) responsive to this improvement in
resolution technologies. For robustness, we test two di�erent de�nitions of systemic importance. For
our �rst test, we isolate all banks that form part of one of the 8 U.S. �nancial holdings that have been
determined 'global systemically important bank' (GSIFI) by the Financial Stability Board.34
As an alternative de�nition, we form a sample of all institutions with asset size larger than USD
50 billion. This is not an entirely arbitrary cuto�, but chosen according to a threshold above which
the Dodd-Frank Act stipulates speci�c enhanced supervision activities and prudential standards, also
in conjunction with the OLA (compare, e.g., DFA, Title II, Sec 210). We use these two de�nitions as
they are alternative, yet not mutually repetitive indicators of systemic importance.35 When we run our
model on these separate samples of banks, we have to use the continuous version of the explanatory
variable since too many institutions would be dropped from the sample otherwise. We are able to
conduct these tests on our bank level sample, with the results being reported in Table 8.
In line with our expectations, the coe�cients of the interaction term emerge to be insigni�cant
for the return volatility as dependent variable in both subsamples. However, it is interesting to note
that for the z-score and asset risk as dependent variable, the coe�cients on the interaction term are
signi�cant, but in opposite direction as compared to our baseline regression results. We interpret this
�nding in a way that more a�ected systemically important banks do not reduce their risk-taking after
the introduction of the OLA, but might even increase it. A possible explanation for this �nding is that
the threat of resolution resulting from the OLA is not credible for them. They do not seem to believe
that the regulator is indeed fully enabled to resolve such institutions in case of failure - be it due to
lacking �nancial or operational capabilities, fears for systemic risk and contagion, or other rationales.
Moreover, as the OLA was considered the major change in bank resolution law in response to the
�nancial crisis, it seems unlikely that these institutions had to expect a further, maybe more credible,
upgrade in resolution technology any time soon. Imagining all �nancial institutions as a system of
corresponding vessels in a situtation where most a�ected institutions have to reduce risk, there are only
few players that can take this risk on - and these are the a�ected institutions for which the resolution
33For clari�cation: The 'a�ected' bank classi�cation is so far not de�ned by size or interconnectedness, but purely ongrounds of resolvability according to the FDIA. Hence, there are, e.g., large as well as small banks being classi�ed as'a�ected' (and 'not a�ected').
34In total, the Financial Stability Board designated 29 institutions to be GSIFI, 8 of which are of U.S. origin: Bankof America, Bank of New York Mellon, Citigroup, Goldman Sachs, JPMorgan Chase, Morgan Stanley, State Street, andWells Fargo.
35Only 24 institutions in our Bank level sample ful�ll both criteria, while additional 40 institutions form part of aGSIFI, and additional 80 institutions report more than USD 50 billion in assets.
31
Table 8: Too-big-to-not-rescue e�ect: Multivariate Di�erence-in-Di�erence analyses on TBTNR banks
(1) (2) (3) (4) (5) (6)Sample Part of U.S.-GSIFI Asset size USD 50+ billionDep. variable Z-score σ RoA Asset risk Z-score σ RoA Asset riskUnregulated share(parent BHC-level) 2.466*** -1.816* 0.721*** 1.133*** -0.892*** 0.111*
(0.948) (0.988) (0.160) (0.367) (0.238) (0.0579)Unregulated sharex after OLA -1.415** 0.0800 0.262*** -0.815* 0.0992 0.0795*
(0.696) (0.295) (0.0643) (0.475) (0.147) (0.0455)
Constant YES YES YES YES YES YESControls YES YES YES YES YES YES
Bank FE YES YES YES YES YES YESTime FE YES YES YES YES YES YES
Observations 485 485 492 452 452 454R-squared 0.824 0.665 0.925 0.863 0.847 0.907
Notes: This table presents multivariate di�erence-in-di�erence estimates of the e�ect that the introduction of the OrderlyLiquidation Authority had on overall risk of those banks that could be classi�ed as too-big-to-not-rescue. The estimation isconducted for two subsamples of banks: All banks that are part of one of the U.S. GSIFIs as classi�ed by the FSB (columns(1) to (3)) and all banks with total asset size of USD 50 billion or more (columns (4) to (6)). A�ected bank takes a value of1 if the bank is part of a BHC with more than 30% of non-FDIA-regulated assets and a value of 0 if the bank is independentor part of a BHC with less than 10% of non-FDIA-regulated assets. After OLA is 1 for the quarters Q3 2010 - Q2 2012and 0 for the quarters Q3 2007 - Q2 2009. Several measures of overall bank risk are taken as dependent variables: z-score(de�ned as mean return on assets plus capital ratio divided by the standard deviation of return on assets), σ RoA (de�nedas standard deviation of return on assets), and asset risk (de�ned as risk-weighted assets divided by total assets). Controlvariables comprise the natural logarithm of total bank assets, capital ratio, pro�tability, liquidity ratio, and an indicatorvariable that takes the value of 1 if the bank was a recipient of the TARP CPP program in a respective quarter (and 0otherwise). All models include bank and time �xed e�ects.
Heteroskedasticity consistent standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
threat is still not credible. Hence, a rational strategy for these 'to-big-to-not-rescue'-institutions would
be to not decrease, but increase risk-taking (at least as long as the resolution threat does not become
more realistic). We cannot test this directly, but the shift in securities and trading asset holdings that
we �nd in the aggregate is at least suggestive to this rationale: While most a�ected institutions that
are not part of a GSIFI heavily reduce their securities holdings (particularly their high-risk securities
and trading assets) after the introduction of the OLA, the a�ected GSIFI institutions even increase
their holdings.
Gambling in the meantime? In a �nal extension, we would like to test how banks' risk-taking
changed in the post-announcement period, i.e. between the proposal of the OLA (mid 2009) and its
actual enactment (mid 2010). Theory and available empirical evidence suggest that gambling might
occur in this period if the changes in regulation reduce a�ected banks' charter value (Fischer et al.,
2012; Murdock et al., 2000). To the extent that the introduction of the OLA actually reduces the
charter value of a�ected banks, e.g. by removing the previously existing implicit bailout guarantee, we
might �nd evidence of gambling in bank behavior. However, banks would need to shift their behavior
twice between the publicly known proposal of the OLA and its signing into law. First, after June 2009,
they would need to increase risk to exploit the trade-o� between high risk-return and potential failure
with likely bailout on the one hand and loss in charter value on the other hand. Very brie�y after this,
before July 2010, bailout becomes less likely (as the OLA becomes e�ective), and banks would need to
readjust their strategy again. Is this realistic? Was the time horizon long enough to shift the behavior
32
twice or was the legislation passed so quickly that gambling did not occur?
In order to test for the occurrence of gambling in the intermediate phase, we de�ne a post-treatment
period (afterOLA) and a 'gambling-period' (afterannouncement) that we run against a pre-treatment
period in turn. While the pre-treatment period is de�ned as Q3 2008 to Q2 2009, the period in which
gambling might happen stretches from Q3 2009 to Q2 2010. For comparison, we de�ne another 4-
quarter post-treatment period as Q3 2010 to Q2 2011 that we use as a benchmark e�ect to compare
it to the gambling results. For robustness, we de�ne an additional set of pre-, gambling-, and post-
treatment periods, which stretch over 2 quarters each: Q1/Q2 2009 as pre-treatment period, Q3/Q4
2009 as potential gambling period, and Q3/Q4 2010 as post-treatment period. We run the main model
with the z-score36 (for overall comparison) as well as a selection of investment choice risk measures that
we deem to be adjustable within a short period, i.e. the trading asset ratio as well as the low and high
risk securities ratios, as dependent variables.37 Panel A in Table 9 presents the results for the 4-quarter
and 2-quarter benchmark regressions (pre- vs post-treatment). It should be noted that these results can
also be interpreted as a robustness test of the initial 8-quarter results. With all overall and investment
risk measures indicating less risk-taking by a�ected banks after the introduction of the OLA in the
4-quarter/2-quarter regressions, these results are fully in line with our baseline model. The �ndings
about potential gambling are displayed in Panel B of Table 9 (pre- vs. gambling-period). Interestingly,
we do not �nd a signi�cant e�ect in the overall risk regression using the z-score as dependent variable.
Likewise, the coe�cient on the interaction term is not signi�cant for the trading assets ratio. However,
the results for low and high risk securities ratios (for the 2-quarter periods only the high-risk securities)
indicate that, if at all, a�ected banks take less - not more - risk in the intermediate period. Interpreting
these �ndings, we do not �nd any evidence for gambling in the intermediate period - rather, a�ected
banks even start decreasing their risk-taking already by shifting their securities portfolio.
How robust are these �ndings? In order to test the robustness of the results presented above,
we have conducted a host of robustness tests, using alternative speci�cations and variable de�nitions,
sample restrictions, and additional entire datasets. This section brie�y summarizes the robustness tests
and their main results. For brevity and ease of comparison, some of the results from the robustness
tests were already presented in the respective tables above. All other results - although not presented
- are also largely consistent with our hypotheses and con�rm the e�ects we report.
The following robustness tests have been carried out:
• With regard to our dependent variables, we have de�ned and tested a set of alternative measures
for overall bank risk and risk choices in business model/investment decisions, both on the bank
level and on the micro-level of business decisions. All of our results have been shown to be robust
to these alterations and yield similar conclusions, indicating that the results are not driven by
speci�c de�nitions of individual dependent variables but largely consistent with each other.
• We acknowledge that the dummy-version of our treatment variable AFFECTEDi is de�ned
along arbitrary cuto�s. In order to test the robustness of our main bank risk-taking results, we
also de�ned alternative cuto�s (0%, 5%, 10% on the lower bound and 30% and 50% on the upper
bound).38 Moreover, we also used the share of non-FDIA-regulated assets as explanatory variable,
36Note that we cannot reasonably de�ne the z-score for the 2-period regressions as it requires the computation of amean and a standard deviation (for which we de�ned a minimum requirement of 3 available datapoints above). Hence,z-score results are only presented for 4-quarter period regressions.
37We tested the two models with all other previously used dependent variables as well and �nd no immediate adjustmente�ect for the intermediate period.
38Concerning the loan level dataset, varying the lower cuto� bound yields similar results. Applying a 50% cuto� for theupper bound is not meaningful as there are only very few banks in the loan level dataset with share of non-FDIA-regulatedassets above this cuto�.
33
Table 9: Bank risk-taking and business model choices: Multivariate Di�erence-in-Di�erence analyses with4-quarter periods and test of risk-taking in post-announcement period
Panel A: Benchmark tests(1) (2) (3) (4) (5) (6) (7)
Level Bank levelPeriods 4-quarter periods 2-quarter periods
Dep. variable Z-score
Tradingassetsratio
Low risksecuritiesratio
High risksecuritiesratio
Tradingassetsratio
Low risksecuritiesratio
High risksecuritiesratio
A�ected bank 0.0889 0.00313*** -0.0240 0.0591** 0.00315 -0.0253 0.125**(0.128) (0.00115) (0.0403) (0.0278) (0.00273) (0.0886) (0.0514)
A�ected bank xafter OLA 0.252*** -0.00568*** 0.0542*** -0.0482*** -0.00390* 0.0517*** -0.0515***
(0.0600) (0.00202) (0.0145) (0.0129) (0.00202) (0.0170) (0.0148)
Constant YES YES YES YES YES YES YESControls YES YES YES YES YES YES YES
Bank FE YES YES YES YES YES YES YESTime FE YES YES YES YES YES YES YES
Observations 28,393 28,579 27,513 21,860 14,597 14,045 11,221R-squared 0.801 0.749 0.850 0.838 0.801 0.892 0.883
Panel B: Gambling tests(1) (2) (3) (4) (5) (6) (7)
Level Bank levelPeriods 4-quarter periods 2-quarter periods
Dep. variable Z-score
Tradingassetsratio
Low risksecuritiesratio
High risksecuritiesratio
Tradingassetsratio
Low risksecuritiesratio
High risksecuritiesratio
A�ected bank 0.0882 -0.000280 -0.0225 0.0131 -0.00269 -0.0493 0.0269**(0.133) (0.00162) (0.0271) (0.0206) (0.00430) (0.0328) (0.0119)
A�ected bank xafter announce-ment -0.00361 0.00285 0.0242** -0.0275** 0.00607 0.00546 -0.0204**
(0.0553) (0.00241) (0.0113) (0.0113) (0.00414) (0.00977) (0.00961)
Constant YES YES YES YES YES YES YESControls YES YES YES YES YES YES YES
Bank FE YES YES YES YES YES YES YESTime FE YES YES YES YES YES YES YES
Observations 29,276 29,472 28,363 22,581 14,653 14,101 11,217R-squared 0.822 0.804 0.900 0.869 0.830 0.951 0.933
Notes: This table presents multivariate di�erence-in-di�erence estimates of the e�ect that the introduction of the Orderly Liq-uidation Authority had on overall bank risk, using pre- and post-treatment periods that stretch over 4 (columns (1) to (4)) or 2(columns (5) to (7)) quarters. Panel A presents benchmark tests comparable to our baseline estimations, but with shorter pre- andpost-treatment periods. Panel B tests the occurence of a gambling e�ect. A�ected bank takes a value of 1 if the bank is part of aBHC with more than 30% of non-FDIA-regulated assets and a value of 0 if the bank is independent or part of a BHC with less than10% of non-FDIA-regulated assets. After OLA is 1 for the quarters Q3 2010 - Q2 2011 (columns (1) to (4)) and Q3 2010 - Q4 2010(columns (5) to (7)) and 0 for the quarters Q3 2008 - Q2 2009 (columns (1) to (4)) and Q1 2009 - Q2 2009 (columns (5) to (7)).After announcement is 1 for the quarters Q3 2009 - Q2 2010 (columns (1) to (4)) and Q3 2009 - Q4 2009 (columns (5) to (7)) and0 for the quarters Q3 2008 - Q2 2009 (columns (1) to (4)) and Q1 2009 - Q2 2009 (columns (5) to (7)). Several dependent variablesare tested: z-score (de�ned as mean return on assets plus capital ratio divided by the standard deviation of return on assets, onlyavailable for the 4-quarter period), trading asset ratio (de�ned as ratio of assets held in trading accounts to total assets), low risksecurities ratio (de�ned as the ratio of securities of U.S. government agencies and subdivisions to total investment securities), andhigh risk securities ratio (de�ned as the ratio of equity securities, asset-backed securities, and trading accounts to total investmentsecurities). Control variables comprise the natural logarithm of total bank assets, capital ratio, pro�tability, liquidity ratio, and anindicator variable that takes the value of 1 if the bank was a recipient of the TARP CPP program in a respective quarter (and 0otherwise). All models include bank and time �xed e�ects.
Heteroskedasticity consistent standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
34
particularly in interaction with afterOLAt. With regard to the de�nition of the treatment period
and the pre- and post-treatment periods, we also use alternative variables, computed over 8, 6,
and 4 quarters. Nevertheless, running our main bank risk-taking model with these alterations
in the key explanatory variables yields results that are comparable in statistical and economic
signi�cance.
• In order to alleviate concerns about endogeneity in our model, we go beyond the univariate
di�erence-in-di�erence approach and add bank and time �xed e�ects for regressions using the
bank level dataset and bank and regional (tract level) �xed e�ects for regressions using the loan
level dataset as well as sets of time-varying control variables (as appropriate). We tested all of
our models in alternative speci�cations, including and excluding the controls and �xed e�ects.
• Where appropriate and mandated by theory, we use alternative model speci�cations. One im-
portant speci�cation is the choice of regression model to test the application approval indicator,
which is a binary variable. In Panel A of Table 7 we presents results using the Linear Prob-
ability Model (LPM) as estimation method. Although the LPM has serious drawbacks (i.e.
heteroskedastic, can predict probabilities outside the range [0;1]), it can be appropriate in a
panel-data setting (see Puri et al. (2011) for a detailed methodological discussion). We rerun
these regressions with probit and logit models and obtain results that are consistent with the
�ndings presented in Table 7.
• Like many other papers using a di�erence-in-di�erence methodology, we rely on a panel dataset
with repeated cross sections of banks and several periods of data before and after the treatment.
Bertrand et al. (2004) describe how this setup can be prone to autocorrelation problems that may
lead to an underestimation of the standard errors. Therefore, we further correct standard errors
for possible autocorrelation at the bank level (as suggested by Puri et al. (2011) and Wooldridge
(2010)) and rerun our models. The results are comparable in size and signi�cance to our �ndings
in the baseline model.
• Finally, we address concerns related to our samples by correcting for outliers, restricting samples
to explanatory variables consistent over time, and using entirely di�erent levels of aggregation.
First, there might be concerns that the results are driven by outliers, e.g. in the dependent
variable or in the non-FDIA-regulated-share that is used to de�ne the treatment variable. In
the bank level dataset, we winsorize the dependent variable, the explanatory variable, and the
control variables with one percent in their highest and lowest quantiles. We run all our tests
using these winsorized versions of dependent, explanatory, and control variables, all together and
each at once. All of our results are robust to these alterations and yield very similar outcomes.
Second, to address concerns about consistency of key explanatory variables, we exclude banks
that change status of AFFECTEDi within our observation period. Our results do not alter
when applying this restriction. Third, we tested our hypotheses on bank risk-taking for di�erent
levels of aggregation: BHC and bank level. Where possible according to data availability, we
test and present both the bank and the BHC level results in parallel, which are largely identical
in direction and signi�cance.
Taken together, our robustness tests suggest that our main �ndings are not driven by variable de�nition,
speci�cation, or sample choice.
35
6 Concluding remarks and policy implications
In July 2010, the U.S. legislator enacted the Orderly Liquidation Authority as part of the �nancial
system reform package, the Dodd-Frank Act. The OLA extends a special bank resolution procedure
to �nancial institutions that were previously not covered by the provisions of the Federal Deposit
Insurance Act, which allows the FDIC to resolve failed banks in an administrative procedure securing
liquidity and discipline alike. Hence, the OLA a�ects �nancial institutions di�erently, raising the
resolution threat particularly for those institutions that were in large part not subject to the FDIA
resolution regime before.
Building on a recent theoretical model by DeYoung et al. (2013), we suggest several hypotheses
how this regulatory change a�ects bank behavior, particularly risk-taking and business model choices.
We propose a di�erence-in-di�erence framework exploiting the di�erential e�ect of the OLA to test
these hypotheses. First and foremost, we �nd the results to be consistent with our main hypothesis:
The introduction of the OLA changes the behavior of the a�ected �nancial institutions towards less
risk-taking and safer business models as compared to the non-a�ected institutions. In the absense of
treatment, i.e. of the regulatory change, both the a�ected and non-a�ected institutions behave equally,
which further corroborates our results. Consistent with the theoretical prediction that the main e�ect
varies with the credibility, capability, and the political will of the regulator to indeed resolve failed
institutions, we �nd the e�ect to vanish for the largest, most systemically relevant institutions. Finally,
we have to reject the hypothesis that a�ected banks gamble in between the announcement and the
enactment of the OLA.
Our �ndings yield several interesting policy implications. If we consider our results an evaluation
of a speci�c change in the U.S. bank resolution regime, we con�rm that the Orderly Liquidation
Authority is indeed an e�ective improvement to the regulatory arsenal. To the extent that a reduction
in overall risk-taking of the previously non-FDIA-regulated �nancial institutions (as compared to their
already regulated peers) was one of the legislator's intentions, our results suggest that it can indeed
be considered successful. However, making OLA's resolution threat credible and thus e�ective for
banks with the highest systemic importance while moderating the liquidity cost of winding down such
institutions will remain a crucial challenge for the regulator.
Moreover, although our analyses focus on the e�ects of a speci�c resolution regime, i.e. the Orderly
Liquidation Authority, our results induce us to also draw general implications for the design or reform
of bank resolution regimes around the world. Based on these �ndings and previous literature, we
propose three fundamental features of e�ective bank resolution regimes that - in our view - can help
to increase and maintain stability in the �nancial system and prevent future �nancial crises. First,
a bank resolution regime that takes into account the special role of �nancial institutions (contrary
to the regular and often inapplicable corporate bankruptcy law) is essential, not only to avoid major
interruptions in liquidity provision, but also to create a credible resolution threat for �nancial institu-
tions in order to discipline them ex ante. Second, comprehensive coverage of �nancial institutions as a
whole - that goes beyond the scope of deposit-taking entities only - will avoid incentives to shift risks
into non-resolvable entities. Third, implementation speed is crucial. When the regulator succeeds in
implementing the resolution threat quickly after its announcement, excessive gambling behavior in the
lag time before enactment can be prevented.
Taken together, a bank resolution regime that incorporates these elements can be more than wishful
thinking - it can be an e�ective threat that disciplines banks towards more prudent behavior.
36
References
Acharya, V., Engle, R., and Pierret, D. (2013). �Testing macroprudential stress tests: The risk of
regulatory risk weights.� Working paper, NYU Stern School of Business.
Acharya, V. V. (2009). �A theory of systemic risk and design of prudential bank regulation.� Journal
of Financial Stability, 5 (3), 224�255.
Acharya, V. V., and Yorulmazer, T. (2007). �Too many to fail�an analysis of time-inconsistency in
bank closure policies.� Journal of Financial Intermediation, 16 (1), 1�31.
Acharya, V. V., and Yorulmazer, T. (2008). �Cash-in-the-market pricing and optimal resolution of
bank failures.� Review of Financial Studies, 21 (6), 2705�2742.
Anand, S. (2011). Essentials of the Dodd-Frank Act. Essentials Series, Wiley.
Ashcraft, A. B. (2005). �Are banks really special? new evidence from the FDIC-induced failure of
healthy banks.� American Economic Review, 95 (5), 1712�1730.
Avery, R. B., Brevoort, K. P., and Canner, G. B. (2007). �Opportunities and issues in using hmda
data.� Journal of Real Estate Research, 29 (4), 351�380.
Bagehot, W. (1873). Lombard Street: A Description of the Money Market. William Clowes and Sons.
Barth, J. R., Caprio, G. J., and Levine, R. (2004). �Bank regulation and supervision: what works
best?� Journal of Financial Intermediation, 13 (2), 205�248.
Berger, A. N., and Bouwman, C. H. (2012). �How does capital a�ect bank performance during �nancial
crises?� Journal of Financial Economics, forthcoming.
Berger, A. N., Bouwman, C. H., Kick, T. K., and Schaeck, K. (2012). �Bank risk taking and liquidity
creation following regulatory interventions and capital support.� Working paper.
Bertrand, M., Du�o, E., and Mullainathan, S. (2004). �How much should we trust di�erences-in-
di�erences estimates?� The Quarterly Journal of Economics, 119 (1), 249�275.
Besanko, D., and Kanatas, G. (1996). �The regulation of bank capital: Do capital standards promote
bank safety?� Journal of Financial Intermediation, 5 (2), 160 � 183.
Black, L. K., and Hazelwood, L. N. (2012). �The e�ect of tarp on bank risk-taking.� Journal of Financial
Stability, (forthcoming).
Bliss, R. R., and Kaufman, G. (2006). �U.s. corporate and bank insolvency regimes: an economic
comparison and evaluation.� Working Paper Series WP-06-01, Federal Reserve Bank of Chicago.
Bliss, R. R., and Kaufman, G. G. (2011). �Resolving insolvent large complex �nancial institutions: A
better way.� The Banking Law Journal, 128 (4), 339�363.
Blum, J. (1999). �Do capital adequacy requirements reduce risks in banking?� Journal of Banking &
Finance, 23 (5), 755 � 771.
Blundell-Wignall, A., and Atkinson, P. (2010). �Thinking beyond Basel III: Necessary solutions for
capital and liquidity.� OECD Journal: Financial Market Trends, 2010 (1), 9�33.
Boyd, J., De Nicolò, G., and Jalal, A. M. (2010). �Bank competition, asset allocations and risk of
failure: An empirical investigation.� CESifo Working Paper Series 3198, CESifo Group Munich.
37
Boyd, J. H., Chang, C., and Smith, B. D. (1998). �Moral hazard under commercial and universal
banking.� Journal of Money, Credit and Banking, 30 (3), 426�468.
Brown, C. O., and Dinç, I. S. (2011). �Too many to fail? evidence of regulatory forbearance when the
banking sector is weak.� Review of Financial Studies, 24 (4), 1378�1405.
Brunnermeier, M. K., Dong, G. N., and Palia, D. (2012). �Banks' non-interest income and systemic
risk.� Working paper series.
Calomiris, C. W. (1999). �Building an incentive-compatible safety net.� Journal of Banking & Finance,
23 (10), 1499 � 1519.
CCH Attorney-Editor (2010). Wall Street Reform and Consumer Protection Act Of 2010: Law, Ex-
planation and Analysis. Wolters Kluwer Law & Business.
Cordella, T., and Yeyati, E. L. (2003). �Bank bailouts: moral hazard vs. value e�ect.� Journal of
Financial Intermediation, 12 (4), 300�330.
Dam, L., and Koetter, M. (2012). �Bank bailouts and moral hazard: Evidence from Germany.� Review
of Financial Studies.
Davies, S. M., and McManus, D. A. (1991). �The e�ects of closure policies on bank risk-taking.� Journal
of Banking & Finance, 15 (4/5), 917 � 938.
DavisPolk (2010). �Summary of the dodd-frank wall street reform and consumer protection act.� Tech.
rep., Davis Polk & Wardwell LLP.
De Nicolò, G., Dell'Ariccia, G., Laeven, L., and Valencia, F. (2010). �Monetary policy and bank risk
taking.� Working Paper Series SPN/10/09, International Monetary Fund.
DeJonghe, O. (2010). �Back to the basics in banking? a micro-analysis of banking system stability.�
Journal of Financial Intermediation, 19 (3), 387�417.
Dell'Ariccia, G., Igan, D., and Laeven, L. (2012). �Credit booms and lending standards: Evidence from
the subprime mortgage market.� Journal of Money, Credit and Banking, 44 (2-3), 367�384.
Demirgüç-Kunt, A., and Detragiache, E. (2002). �Does deposit insurance increase banking system
stability? an empirical investigation.� Journal of Monetary Economics, 49 (7), 1373 � 1406.
Demirgüç-Kunt, A., and Huizinga, H. (2004). �Market discipline and deposit insurance.� Journal of
Monetary Economics, 51 (2), 375 � 399.
Demirgüç-Kunt, A., and Huizinga, H. (2010). �Bank activity and funding strategies: The impact on
risk and returns.� Journal of Financial Economics, 98 (3), 626�650.
Department of the Treasury (2009). �A new foundation : rebuilding �nancial supervision and regula-
tion.� Tech. rep.
DeYoung, R. (2013). �Modelling economies of scale in banking: Simple versus complex models.� In
F. Pasiouras (Ed.), E�ciency and Productivity Growth: Modelling in the Financial Services Industry,
chap. 3, Wiley.
DeYoung, R., Kowalik, M., and Reidhill, J. (2013). �A theory of failed bank resolution: Technological
change and political economics.� Journal of Financial Stability, forthcoming.
38
DeYoung, R., and Roland, K. P. (2001). �Product mix and earnings volatility at commercial banks:
Evidence from a degree of total leverage model.� Journal of Financial Intermediation, 10 (1), 54 �
84.
Diamond, D. W., and Dybvig, P. H. (1983). �Bank runs, deposit insurance, and liquidity.� Journal of
Political Economy, 91 (3), 401�19.
Diamond, D. W., and Rajan, R. G. (2005). �Liquidity shortages and banking crises.� The Journal of
Finance, 60 (2), 615�647.
Duchin, R., and Sosyura, D. (2012). �Safer ratios, riskier portfolios: Banks' response to government
aid.� Working Paper 1165, Ross School of Business.
Edwards, J. M. (2011). �Fdicia v. dodd-frank: Unlearned lessons about regulatory forbearance?� Har-
vard Business Law Review, 1, 279�301.
European Commission (2012). �Communication from the commission to the european parliament and
the council: A roadmap towards a banking union.� Communication COM(2012) 510 �nal, European
Commission.
FDIC (2011). �The orderly liquidation of lehman brothers holdings inc. under the dodd-frank act.�
FDIC Quarterly, 5 (2), 31�49.
Fischer, M., Hainz, C., Rocholl, J., and Ste�en, S. (2012). �Government guarantees and bank risk
taking incentives.� Working paper.
Fitzpatrick, T. J., Kearney-Marks, M., and Thomson, J. B. (2012). �The history and rationale for a
separate bank resolution process.� Economic Commentary, (1).
Gropp, R., Gruendl, C., and Guettler, A. (2010). �The impact of public guarantees on bank risk taking:
evidence from a natural experiment.� Working Paper Series 1272, European Central Bank.
Gropp, R., Hakenes, H., and Schnabel, I. (2011). �Competition, risk-shifting, and public bail-out
policies.� Review of Financial Studies, 24 (6), 2084�2120.
Gropp, R., and Vesala, J. (2004). �Deposit insurance, moral hazard and market monitoring.� Review
of Finance, 8 (4), 571�602.
Gropp, R., Vesala, J., and Vulpes, G. (2006). �Equity and bond market signals as leading indicators
of bank fragility.� Journal of Money, Credit and Banking, 38 (2), 399�428.
Hakenes, H., and Schnabel, I. (2010). �Banks without parachutes: Competitive e�ects of government
bail-out policies.� Journal of Financial Stability, 6 (3), 156 � 168.
Kareken, J. H., and Wallace, N. (1978). �Deposit insurance and bank regulation: A partial-equilibrium
exposition.� The Journal of Business, 51 (3), 413�438.
Kasa, K., and Spiegel, M. M. (2008). �The role of relative performance in bank closure decisions.�
Federal Reserve Bank of San Francisco Economic Review, 17�29.
Keeley, M. C. (1990). �Deposit insurance, risk, and market power in banking.� The American Economic
Review, 80 (5), 1183�1200.
Kim, D., and Santomero, A. M. (1988). �Risk in banking and capital regulation.� Journal of Finance,
43 (5), 1219 � 1233.
39
Koehn, M., and Santomero, A. M. (1980). �Regulation of bank capital and portfolio risk.� Journal of
Finance, 35 (5), 1235 � 1244.
Laeven, L., and Levine, R. (2009). �Bank governance, regulation and risk taking.� Journal of Financial
Economics, 93 (2), 259�275.
Loutskina, E., and Strahan, P. E. (2009). �Securitization and the declining impact of bank �nance on
loan supply: Evidence from mortgage originations.� The Journal of Finance, 64 (2), 861�889.
Mailath, G. J., and Mester, L. J. (1994). �A positive analysis of bank closure.� Journal of Financial
Intermediation, 3 (3), 272�299.
Marin, M., and Vlahu, R. (2011). The Economics of Bank Bankruptcy Law. Springer.
Merton, R. C. (1977). �An analytic derivation of the cost of deposit insurance and loan guarantees an
application of modern option pricing theory.� Journal of Banking & Finance, 1 (1), 3 � 11.
Murdock, K. C., Hellmann, T. F., and Stiglitz, J. E. (2000). �Liberalization, moral hazard in banking,
and prudential regulation: Are capital requirements enough?� American Economic Review, 90 (1),
147�165.
Perotti, E. C., and Suarez, J. (2002). �Last bank standing: What do I gain if you fail?� European
Economic Review, 46 (9), 1599�1622.
Puri, M., Rocholl, J., and Ste�en, S. (2011). �Global retail lending in the aftermath of the us �nancial
crisis: Distinguishing between supply and demand e�ects.� Journal of Financial Economics, 100 (3),
556�578.
Repullo, R. (2004). �Capital requirements, market power, and risk-taking in banking.� Journal of
Financial Intermediation, 13 (2), 156 � 182, bank Capital Adequacy Regulation under the New
Basel Accord.
Rosen, R. J. (2011). �Competition in mortgage markets: the e�ect of lender type on loan characteris-
tics.� Economic Perspectives, 35 (1), 2�21.
Roy, A. D. (1952). �Safety �rst and the holding of assets.� Econometrica, 20 (3), 431�449.
Scott, K., Shultz, G., and Taylor, J. (Eds.) (2010). Ending Government Bailouts As We Know Them.
No. 1 in Books, Hoover Institution, Stanford University.
Scott, K. E., and Taylor, J. B. (Eds.) (2012). Bankruptcy Not Bailout. No. 5 in Books, Hoover Insti-
tution, Stanford University.
Sironi, A. (2003). �Testing for market discipline in the european banking industry: Evidence from
subordinated debt issues.� Journal of Money, Credit & Banking (Ohio State University Press),
35 (3), 443 � 472.
Thakor, A. V. (1996). �Capital requirements, monetary policy, and aggregate bank lending: Theory
and empirical evidence.� The Journal of Finance, 51 (1), 279�324.
Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data. MIT Press.
Zaring, D. T. (2010). �A lack of resolution.� Emory Law Journal, 60 (1), 97�157.
40